METHOD AND SYSTEM FOR IMAGE PROCESSING
FIELD
[0001] Various embodiments of the disclosure relate to a method and system for
image processing. More specifically, various embodiments of the disclosure relate
to a method and system for image processing, to determine a region of interest.
BACKGROUND
[0002] In image processing, conversion of two-dimensional (2D) images to
three-dimensional (3D) images is a growing area of interest in a variety of
applications. Relative depth assignment to objects in a 3D image plays an
important role in 2D-to-3D image conversion. Depth of objects in an image is
assigned in relation to a background. To accomplish this, the image is segmented
to distinguish these objects from the background. However, certain objects in the
image, such as human hair, differ significantly in appearance from person-toperson.
Differences may include varying tones of hair color and lighting effects. In
occasional cases, hair and background colors may have similar intensities. In view
of these challenges, existing techniques fail to accurately distinguish certain objects
from the background, and therefore fail to assign correct relative depth to these
objects.
[0003] Further limitations and disadvantages of conventional and traditional
approaches will become apparent to one of skill in the art, through comparison of
such systems with some aspects of the present disclosure, as set forth in the
remainder of the present application, with reference to the drawings.
SUMMARY
[0004] A system and method are provided for image processing substantially as
shown in, and/or described in connection with, at least one of the figures, as set
forth more completely in the claims.
[0005] These and other features and advantages of the present disclosure may
be appreciated from a review of the following detailed description of the present
disclosure, along with the accompanying figures in which like reference numerals
refer to like parts throughout.
BRIEF DESCRIPTION OF THE DRAWINGS
[0006] FIG. 1 is a block diagram illustrating a network environment for image
processing, in accordance with an embodiment of the disclosure.
[0007] FIG. 2 is a block diagram illustrating a computing device for image
processing to determine a subject region and a background region, in accordance
with an embodiment of the disclosure.
[0008] FIG. 3 is a diagram illustrating a method for image processing to
determine a subject region and a background region, in accordance with an
embodiment of the disclosure.
[0009] FIG. 4 is a diagram illustrating a method for image processing to
determine a region of interest, in accordance with an embodiment of the disclosure.
[0010] FIG. 5 is a flow chart illustrating a sequence of sub-tasks for hair
segmentation, in accordance with an embodiment of the disclosure.
[001 1] FIGS. 6A-6B are diagrams illustrating screenshots of a two-dimensional
input image and a down-quantized image, in accordance with an embodiment of
the disclosure.
[0012] FIGS. 7A-7B are diagrams illustrating screenshots of a two-dimensional
input image and a cropped, resized and down-quantized image, in accordance with
an embodiment of the disclosure.
[0013] FIG. 8 is a diagram illustrating a screenshot of a two-dimensional image
showing predicted locations for different classes of pixels, in accordance with an
embodiment of the disclosure.
[0014] FIGS. 9A-9B are diagrams illustrating histogram plots for hair and
background areas of a two-dimensional image, in accordance with an embodiment
of the disclosure.
[0015] FIG. 10 is a diagram illustrating gradient-based distinctiveness functions
for hair class as compared to background class, in accordance with an embodiment
of the disclosure.
[0016] FIG. 11 is a diagram illustrating a screenshot of a Boolean-type image
showing the region of interest and a background region, in accordance with an
embodiment of the disclosure.
[0017] FIG. 12 is a diagram illustrating a method for post-processing of the
determined region of interest, in accordance with an embodiment of the disclosure.
[0018] FIG. 13A is a diagram illustrating a screenshot of a smoothed image of
the region of interest, in accordance with an embodiment of the disclosure.
[0019] FIG. 13B is a diagram illustrating screenshots of a smoothed image of
the region of interest before and after eliminating outlier patches, in accordance
with an embodiment of the disclosure.
[0020] FIG. 13C is a diagram illustrating a screenshot of a region of interest
mask, in accordance with an embodiment of the disclosure.
[0021] FIG. 13D is a diagram illustrating a screenshot of the mask image
superimposed on the two-dimensional input image, in accordance with an
embodiment of the disclosure.
DETAILED DESCRIPTION
[0022] Various implementations may be found in a system and method for
image processing to determine a region of interest.
[0023] Exemplary aspects of a method for image processing may include a
computing device having one or more processors. In an embodiment, the
computing device may be operable to determine luminance values of multiple pixels
in a subset of a frame of a two-dimensional image. The computing device is further
operable to determining texture values of the multiple pixels in the subset of the
frame. The computing device is operable to identify a subject region and a
background region in the frame of the two-dimensional image based on the
determined luminance values and the determined texture values of the multiple
pixels.
[0024] In an embodiment, the luminance values and the textures values of the
pixels are stored in a first set of lookup tables and a second set of lookup tables
respectively. The identification of the subject region and the background region in
the frame of the two-dimensional image comprises classifying the multiple pixels
into a first class and/or a second class based on a comparison of the luminance
values and the texture values of the pixels with the first set of lookup tables and the
second set of lookup tables respectively. In an embodiment, the first class
comprises a head class and the second class comprises a background class. The
head class may further comprise a face class and a hair class.
[0025] In an embodiment, the computing device is further operable to convert
the two-dimensional image to a three-dimensional image based on the identification
of the subject region and the background region in the two-dimensional image.
[0026] Exemplary aspects of a method for image processing may include a
computing device having one or more processors. In an embodiment, the
computing device may be operable to generate a first set of values, based on
luminance of a first class of pixels and a second class of pixels, in a first twodimensional
image. As used herein, the luminance of pixels in an image is the
brightness distribution or luminosity information of one or more objects in the
image. The computing device is further operable to generate a second set of
values, based on texture of the first class of pixels and the second class of pixels in
the first two-dimensional image. The computing device may classify a subject pixel
into one of a first class of pixels and a second class of pixels based on a
comparison of a luminance value of the subject pixel and a texture value of the
subject pixel with the generated first set of values and the generated second set of
values respectively.
[0027] In an embodiment, the subject pixel belongs to the first two dimensional
image. In another embodiment, the subject pixel belongs to a second twodimensional
image different from the first two-dimensional image. In an
embodiment, the first class comprises a head class and the second class
comprises a background class. The head class may further comprise a face class
and a hair class.
[0028] In an embodiment, the computing device may determine a region of
interest in the first two-dimensional image, based on the first set of values and the
second set of values. In an embodiment, the computing device may determine a
region of interest in a second two-dimensional image, based on the first set of
values and the second set of values.
[0029] The first set of values corresponds to a first set of lookup tables (LUTs)
that store the histogram counts of luminance values from the first class of pixels
and the second class of pixels of the first two-dimensional image. The first set of
LUTs represent class conditional probability distributions for the first and second
class of pixels. The second set of values corresponds to a second set of LUTs to
store gradient distinctiveness indicator functions for the first class of pixels.
[0030] Exemplary aspects of a method for image processing may include a
computing device having one or more processors. In an embodiment, the
computing device may be operable to determine a region of interest (ROI) in a twodimensional
image, herein referred to as a two-dimensional ROI image. The
computing device may determine a first class of pixels and a second class of pixels
in the two-dimensional ROI image based on a first set of predetermined values
corresponding to a luminance of the two-dimensional image. The computing device
may analyze the texture of the first class of pixels and the second class of pixels
corresponding to the texture of said two-dimensional image to determine the region
of interest. The computing device may be further operable to convert the twodimensional
image to a three-dimensional image based on the determination of the
region of interest in the two-dimensional ROI image. The 3D images obtained from
the conversion are then utilized to render 3D media content.
[0031] In an embodiment, the two-dimensional ROI image is generated from a
two-dimensional input image. In an example, a two-dimensional (2D) media content
comprising a sequence of moving images is divided into multiple 2D image frames.
Each image frame from the sequence of moving images represents the twodimensional
input image. The two-dimensional ROI image is generated by
cropping, resizing and down-quantizing the two-dimensional input image. The twodimensional
ROI image has a lower dimension compared to the two-dimensional
input image.
[0032] In an embodiment, the method further comprises predicting a first region
and a second region in the two-dimensional ROI image. The predicted first region
and second region correspond to sample regions in the two-dimensional ROI image
where the first class of pixels and second class of pixels are likely to be present.
For example, the predicted first region and the predicted second region comprise a
head region and a background region, respectively. The first class of pixels and
second class of pixels are determined by computing class-conditional probability
functions using Bayes rule, based on luminance of the first and second class of
pixels and a first set of predetermined values. Each class conditional probability
function represents the probability of one event on the condition that another event
occurs.
[0033] In an embodiment, the first set of predetermined values corresponds to a
first set of lookup tables (LUTs). The first set of LUTs store histogram counts of
luminance values from the first class of pixels and the second class of pixels of the
two-dimensional ROI image and/or at least one previous two-dimensional image.
[0034] The analyses of texture comprise computing gradient distinctiveness
indicator functions for the first class of pixels and the second class of pixels. The
uniqueness of the first class of pixels as compared to the second class of pixels is
determined based on the computed gradient distinctiveness indicator functions and
a second set of predetermined values. In an embodiment, the second set of
predetermined values corresponds to a second set of lookup tables. The second
set of lookup tables store the gradient distinctiveness indicator functions for the first
class of pixels of the two-dimensional ROI image and/or at least one previous twodimensional
image.
[0035] In an embodiment, the first class of pixels comprises a head class and
the second class of pixels comprises a background class. The head class may
further comprise a face class and a hair class. The region of interest is a human
hair region, a head region, or a combination thereof. In an embodiment, the
determined region of interest is represented as a Boolean-type image.
[0036] In an embodiment, the region of interest represented as a Boolean-type
image is subjected to one or more post-processing tasks. The region of interest is
subjected to image smoothing by filling gaps within the region of interest. The
disconnected patches, or outliers, around the region of interest are then eliminated.
The outer contour of the region of interest is detected to generate a region of
interest (ROI) image mask. The ROI image mask is then restored to the original
scale of the two-dimensional input image.
[0037] FIG. 1 is a block diagram illustrating a network environment for image
processing, in accordance with an embodiment of the disclosure. With reference to
FIG. 1, there is shown a network environment 100. The network environment 100
may comprise a server computing device 102, a user computing device 104, a
media management device 106, a database 108, and a communication network
110. The server computing device 102 and the user computing device 104 may
also be collectively referred to herein as the computing device 102/104.
[0038] The server computing device 102 may comprise suitable logic, circuitry,
interfaces, and/or code that may enable receiving and processing of still and
moving images directly from the media management device 106, or via the
communication network 110. For example, the server computing device 102 may
be operable to perform image segmentation of the received images, relative depth
assignment to objects in the images, and 2D-to-3D image conversion. In an
embodiment, the server computing device 102 may be implemented as a cloudbased
server. The server computing device 102 may store and periodically update
one or more sets of self-learning LUTs, either autonomously or in association with
the database 108. An administrator user may interact with the server computing
device 102 directly. In other embodiments, the administrator user may issue
commands or instructions indirectly to the server computing device 102 over the
communication network 110 .
[0039] The user computing device 104 may comprise suitable logic, circuitry,
interfaces, and/or code that may enable receiving and processing of still and
moving images directly from the media management device 106, or via the
communication network 110 . For example, the user computing device 104 may also
be operable to perform image segmentation of the received images, relative depth
assignment to objects in the images, and 2D-to-3D image conversion. The user
computing device 104 may include one or more of a computer, a smart phone, a
mobile device, a personal digital assistant (PDA), a laptop, a tablet PC, and
consumer devices or appliances, such as a television and/or a set-top box.
Notwithstanding, the disclosure may not be so limited, and other types of user
computing devices 104 may be communicatively coupled to the media
management device 106 and the server computing device 102, without limiting the
scope of the disclosure. The user computing device 108 may also include
computing devices capable of receiving and rendering media content. The user
computing device 104 may also be operable to store and periodically update one or
more sets of self-learning LUTs in association with the database 108. An end user
may interact with the user computing device 104 directly. In other embodiments, the
administrator user may issue commands or instructions indirectly to the server
computing device 102 over the communication network 110 .
[0040] The media management device 106 may comprise suitable logic,
circuitry, interfaces, and/or code that may enable capturing, storing, and distributing
media directly to the server computing device 102 and/or the user computing
device 104, or via the communication network 110 . For example, the captured
media comprises still images and moving images. The media management device
106 may be operable to capture the moving and/or still images at configurable
resolutions and in different image-capture formats. The media management device
106 may be further operable to perform production, distribution and management of
various forms of media content such as television content, on-demand media
content, web-based content, and personal content. The format of the media content
can include, but not limited to, AVI, MPEG, and DivX. In an embodiment, the media
management device 106 may be operable to broadcast moving images to user
computing devices 104 via the network. The media management device 106 may
distribute or transfer the moving images to the server computing device 102, or the
user computing device 104, via wired or wireless communication protocols. The
media management device 106 may also comprise mass storage devices to store
the captured media for subsequent transfer or distribution.
[0041] The database 108 may comprise suitable logic, circuitry, interfaces,
and/or code that may enable storage and transfer of media and data to the server
computing device 102 and the user computing device 104. The database 108 is
operable to store and periodically update one or more sets of self-learning LUTs,
according to instructions from the server computing device 102 and the user
computing device 104.
[0042] The communication network 110 may include a medium through which
various computing devices 104, in the computing environment 100, may receive
and transmit media content and data. Examples of the communication network 110
may include, but are not limited to, the Internet, a Wireless Fidelity (WiFi) network,
a Wireless Local Area Network (WLAN), a Local Area Network (LAN), a telephone
line (POTS), or a Metropolitan Area Network (MAN). Various devices in the
computing environment 100 may be operable to connect to the communication
network 110, in accordance with various wired and wireless communication
protocols, such as Transmission Control Protocol and Internet Protocol (TCP/IP),
User Datagram Protocol (UDP), Hypertext Transfer Protocol (HTTP), File Transfer
Protocol (FTP), ZigBee, EDGE, Infra Red (IR), IEEE 802.1 1, 802.16, cellular
communication protocols, and/or Bluetooth (BT) communication protocols.
[0043] In operation, the server computing device 102 or the user computing
device 104, collectively referred to herein as the computing device 102/1 04, may
receive two-dimensional (2D) media content such as moving images or still images
from the media management device 106. In an embodiment, the computing device
102/1 04 may divide the moving images to create a sequence of individual 2D
image frames in a suitable format, including but not limited to, joint photographic
experts group (JPEG), portable network graphics (PNG), graphics interchange
format (GIF), and tagged image file format (TIFF) formats. The computing device
102/1 04 may generate luminance information from the 2D image frames. The
computing device 102/1 04 may extract the luminance information from the 2D
image frames. In an embodiment, the computing device 102/1 04 may receive twodimensional
images directly from the media management device 106, or from
another computing device. The computing device 102/1 04 may determine
luminance values of multiple pixels in a subset of a frame of a two-dimensional
image. The computing device 102/1 04 may determine texture values of the multiple
pixels in the subset of the frame based on distinctive gradients of the pixels. The
computing device 102/1 04 may identify a subject region and a background region
in the frame of the two-dimensional image based on the determined luminance
values and the determined texture values of the multiple pixels.
[0044] In an embodiment, the computing device 102/1 04 may identify the
subject region and the background region in the frame of the two-dimensional
image by classifying the pixels into a first class and/or a second class based on a
comparison of the luminance values and the texture values of the pixels with a first
set of lookup tables and a second set of lookup tables respectively. In an example,
the identified subject region comprises a hair region and/or a head region.
[0045] In an embodiment, the computing device 102/1 04 may convert the twodimensional
image to a three-dimensional image based on the identification of the
subject region and the background region in the two-dimensional image.
[0046] In an embodiment, the computing device 102/1 04 may generate a first
set of values based on luminance of a first class of pixels and a second class of
pixels in a first two-dimensional image. In an embodiment, the first set of values
corresponds to a first set of lookup tables (LUTs). The first set of LUTs store
histogram counts of luminance values for a first class of pixels and a second class
of pixels. The first set of LUTs represent class conditional probability distributions
for the first and second class of pixels. The computing device 102/104 may
generate a second set of values based on the texture of the first class of pixels and
the second class of pixels in the first two-dimensional image. The second set of
values corresponds to a second set of LUTs. In an embodiment, the second set of
LUTs store gradient distinctiveness indicator functions for the first class of pixels. In
an embodiment, the computing device 102/1 04 may generate the first set of LUTs
and the second set of LUTs as part of a continuous self-training routine. The
computing device 102/1 04 may transfer the generated first set of LUTs and the
second set of LUTs to the database 108 for storage. The computing device
102/1 04 may determine a region of interest in the first two-dimensional image
based on the first set of values and the second set of values. In an embodiment,
the computing device 102/1 04 may determine a region of interest in a second twodimensional
image based on the first set of values and the second set of values.
[0047] In an embodiment, the computing device 102/1 04 may determine a
region of interest (ROI) in a two-dimensional image, herein also referred to as a
two-dimensional ROI image. The computing device 102/1 04 may determine a first
class of pixels and a second class of pixels in the two-dimensional ROI image
based on a first set of predetermined values. For example, the first class of pixels
comprises a head class and the second class of pixels comprises a background
class. In an embodiment, the head class further comprises a face class and a hair
class. The computing device 102/1 04 may analyze the texture of the first class of
pixels and the second class of pixels based on a second set of predetermined
values. The determined region of interest comprises a head region, a hair region, or
a combination thereof.
[0048] In an embodiment, the computing device 102/1 04 may be operable to
predict a first region and a second region in the two-dimensional ROI image. The
predicted first region and second region correspond to sample regions in the twodimensional
ROI image where the first class of pixels and second class of pixels
are likely to be present. For example, the predicted first region and the predicted
second region may comprise a head region and a background region, respectively.
The computing device 102/1 04 may determine the first class of pixels and second
class of pixels by computing class-conditional probability functions using Bayes rule
based on luminance of the first and second class of pixels and the first set of
predetermined values.
[0049] In an embodiment, the first set of predetermined values corresponds to a
first set of lookup tables (LUTs). The first set of LUTs store histogram counts of
luminance values of the first class of pixels and the second class of pixels of the
two-dimensional ROI image and/or at least one previous two-dimensional image.
[0050] In an embodiment, the computing device 102/1 04 may be operable to
analyze the texture by computing gradient distinctiveness indicator functions for the
first class of pixels and the second class of pixels. The computing device 102/1 04
determines the uniqueness of the first class of pixels as compared to the second
class of pixels, based on the computed gradient distinctiveness indicator functions
and a second set of predetermined values. In an embodiment, the second set of
predetermined values corresponds to a second set of LUTs. The second set of
LUTs store the gradient distinctiveness indicator functions for the first class of
pixels from the two-dimensional ROI image and/or at least one previous twodimensional
image.
[0051] In an embodiment, the computing device 102/1 04 may generate the twodimensional
ROI image from a two-dimensional input image. Each two-dimensional
input image is a two-dimensional (2D) image frame from a sequence of moving
images of a 2D media content. The computing device 102/1 04 generates the twodimensional
ROI image by cropping, resizing and down quantizing each twodimensional
input image. The order and recurrence of the cropping, resizing and
down-quantizing operations to generate the two-dimensional ROI image may be
predetermined or random. The two-dimensional ROI image has a lower dimension
compared to the two-dimensional input image. In an embodiment, the computing
device 102/1 04 may be operable to convert each 2D image frame to a 3D image,
based on the determination of the region of interest in the 2D image frame.
[0052] FIG. 2 is a block diagram illustrating a computing device for image
processing to determine a region of interest, in accordance with an embodiment of
the disclosure. FIG. 2 is explained in conjunction with elements from FIG. 1. With
reference to FIG. 2, there is shown the computing device 102/1 04. In an
embodiment, the computing device depicted in FIG. 2 is the server computing
device 102. In another embodiment, the computing device depicted in FIG. 2 is the
user computing device 104. The computing device 102/1 04 may comprise a
processor 202, a memory 204, Input-Output (I/O) devices 206, and a transceiver
208.
[0053] The processor 202 may be communicatively coupled to the memory 204,
and the I/O devices 206. Further, the transceiver 208 may be communicatively
coupled to the processor 202, the memory 204, and the I/O devices 206.
[0054] The processor 202 may comprise suitable logic, circuitry, interfaces,
and/or code that may be operable to execute a set of instructions stored in the
memory 204. The processor 202 may be implemented based on a number of
processor technologies known in the art. Examples of the processor 202 may be an
X86-based processor, a RISC processor, an ASIC processor, a CISC processor, or
any other processor.
[0055] The memory 204 may comprise suitable logic, circuitry, interfaces, and/or
code that may be operable to store the received set of instructions. The memory
204 may be implemented based on, but not limited to, a Random Access Memory
(RAM), a Read-Only Memory (ROM), a Hard Disk Drive (HDD), a storage server
and/or a secure digital (SD) card.
[0056] The I/O devices 206 may comprise various input and output devices
operably connected to the processor 202. Examples of input devices include, but
are not limited to, a keyboard, a mouse, a joystick, a touch screen, a microphone, a
camera, a motion sensor, a light sensor, and/or a docking station. Examples of
output devices include, but are not limited to, a display and a speaker.
[0057] The transceiver 208 may comprise suitable logic, circuitry, interfaces,
and/or code that may be operable to communicate with the media management
device 106 and the database 108. The transceiver 208 may implement known
technologies for supporting wired or wireless communication with the
communication network 110 .
[0058] In operation, the processor 202 may be operable to receive twodimensional
(2D) media content such as moving images or still images from the
media management device 106. In an embodiment, the processor 202 may divide
the moving images to create a sequence of individual 2D image frames in a
suitable format. The processor 202 may generate luminance-only images from the
2D image frames. The processor 202 may generate the luminance-only images by
extracting luminance information from the 2D image frames. Each 2D image frame
constitutes a two-dimensional input image to the processor 202. In an embodiment,
the processor 202 may generate a two-dimensional ROI image of a lower
dimension from the two-dimensional input image. The two-dimensional ROI image
is generated by cropping, resizing and down-quantizing the two-dimensional input
image, based on an object bounding box of an object-detection algorithm. The
order and recurrence of the cropping, resizing and down-quantizing operations to
generate the two-dimensional ROI image may be predetermined or arbitrary. The
memory requirement of the two-dimensional ROI image is lower than the memory
requirement of the two-dimensional input image.
[0059] The processor 202 may be operable to determine luminance values of
multiple pixels in a subset of a frame of a two-dimensional ROI image. The
processor 202 may be operable to determine texture values of the multiple pixels in
the subset of the frame based on distinctive gradients of the pixels. The processor
202 may be operable to identify a subject region and a background region in the
frame of the two-dimensional image based on the determined luminance values
and the determined texture values of the multiple pixels.
[0060] In an embodiment, the processor 202 may be operable to identify the
subject region and the background region in the frame of the two-dimensional ROI
image by classifying the pixels into a first class and/or a second class based on a
comparison of the luminance values and the texture values of the pixels with a first
set of lookup tables and a second set of lookup tables respectively. In an example,
the identified subject region comprises a hair region and/or a face region.
[0061] In an embodiment, the processor 202 may be operable to convert the
two-dimensional image to a three-dimensional image based on the identification of
the subject region and the background region in the two-dimensional ROI image.
[0062] In an embodiment, the processor 202 may be operable to generate a first
set of values, based on luminance of a first class of pixels and a second class of
pixels in a first two-dimensional ROI image. For example, the first set of values
corresponds to a first set of lookup tables (LUTs) to store histogram counts of
luminance values for a first class of pixels and a second class of pixels. The first set
of LUTs represent class conditional probability distributions for the first and second
class of pixels. The processor 202 may generate a second set of values based on
the texture of the first class of pixels and the second class of pixels in the first twodimensional
ROI image. For example, the second set of values corresponds to a
second set of LUTs to store gradient distinctiveness indicator functions for the first
class of pixels in the first two-dimensional ROI image. In an embodiment, the
processor 202 may generate the first set of LUTs and the second set of LUTs as
part of a self-training routine. The processor 202 may transfer the generated first
set of LUTs and the second set of LUTs to the database 108 for storage.
[0063] The processor 202 may determine a region of interest (ROI) in the first
two-dimensional ROI image based on the first set of values and the second set of
values. In an embodiment, the processor 202 determines a region of interest in a
second two-dimensional ROI image, based on the first set of values and the
second set of values. The processor 202 may determine a first class of pixels and a
second class of pixels in the second two-dimensional ROI image, based on the first
set of values. For example, the first class of pixels comprises a head class and the
second class of pixels comprises a background class. In an embodiment, the head
class further comprises a face class and a hair class. The processor 202 may
analyze the texture of the first class of pixels and the second class of pixels in the
second two-dimensional ROI image, based on the second set of values. The
determined region of interest comprises a head region, a hair region, or a
combination thereof. The image processing to determine the region of interest, in
accordance with embodiments of the disclosure, may be performed either at the
server computing device 102 or at the user computing device 104.
[0064] In an embodiment, the processor 202 may be operable to predict a first
region and a second region in the second two-dimensional ROI image. The
predicted first region and second region correspond to exemplar regions in the
second two-dimensional ROI image where the first class of pixels and second class
of pixels are likely to be present. For example, the predicted first region and the
predicted second region comprise a head region and a background region,
respectively. The processor 202 may determine the first class of pixels and second
class of pixels in the second luminance by computing class-conditional probability
functions, based on luminance of the first and second class of pixels and Bayes
rule.
[0065] In an embodiment, the processor 202 may be operable to analyze the
texture by computing gradient distinctiveness indicator functions for the first class of
pixels and the second class of pixels. The processor 202 may determine the
uniqueness of the first class of pixels as compared to the second class of pixels in
the second two-dimensional ROI image, based on the gradient distinctiveness
indicator functions and the second set of values.
[0066] In an embodiment, the computing device converts 2D images to 3D
images based on the determination of the region of interest. The 2D-to-3D
conversion can be performed either at the server computing device 102 or at the
user computing device 104, and both are within the scope of the disclosure.
[0067] FIG. 3 is a diagram illustrating a method 300 for image processing to
determine a subject region and a background region, in accordance with an
embodiment of the disclosure. FIG. 3 will be described in conjunction with elements
of FIG. 1 and FIG. 2 . The method 300 may be implemented in a computing device
102/1 04 comprising one or more processors 202.
[0068] The method 300 starts at 302. At step 304, the luminance values of
multiple pixels in a subset of a frame of a two-dimensional image may be
determined by the computing device 102/104. At step 306, texture values of the
multiple pixels in the subset of the frame may be determined by the computing
device 102/1 04, based on distinctive gradients of the pixels. The luminance values
and the textures values of the multiple pixels are stored in a first set of lookup
tables and a second set of lookup tables respectively.
[0069] At step 308, a subject region and a background region in the frame of the
two-dimensional image is identified based on the determined luminance values and
the determined texture values of the pixels. The identification of the subject region
and the background region in the frame of the two-dimensional image may
comprise classifying the plurality of pixels into a first class and/or a second class
based on a comparison of the luminance values and the texture values of the pixels
with a first set of lookup tables and a second set of lookup tables respectively. At
step 3 10, the method 300 ends.
[0070] FIG. 4 is a diagram illustrating a method 400 for image processing to
determine a region of interest, in accordance with an embodiment of the disclosure.
FIG. 4 will be described in conjunction with elements of FIG. 1 and FIG. 2. The
method 400 may be implemented in a computing device 102/104 comprising one or
more processors 202.
[0071] The method 400 starts at 402. At step 404, a two-dimensional ROI image
is generated from a two-dimensional input image by the computing device 102/1 04,
based on an object bounding box. For example, the object-bounding box is a facedetection-
based face bounding box. The two-dimensional input image is cropped,
resized and down-quantized, based on the face bounding box. In order to reduce
the memory and gate usage for the hardware implementation, the computing
device 102/1 04 crops, resizes, and down quantizes a broader face two-dimensional
input image to a smaller two-dimensional ROI image. The memory requirement of
the smaller two-dimensional ROI image is lower than the memory requirement of
the actual two-dimensional input image. In an embodiment, the computing device
102/1 04 may perform one or more of the cropping, resizing and down quantizing
operations on the two-dimensional input image to generate the smaller twodimensional
ROI image.
[0072] At step 406, a first set of values may be generated by the computing
device 102/104, based on luminance values of pixels in the two-dimensional ROI
image. In an embodiment, the computing device 102/1 04 processes one or more
two-dimensional images using a frame-by-frame-based self-training algorithm that
uses luminance-only image information. The computing device 102/1 04 may collect
positive and negative exemplars with reference to the face-bounding box to form
probabilistic and discriminant functions for hair and background regions. In an
embodiment, the first set of values corresponds to a first set of lookup tables
(LUTs) that store histogram counts of luminance values for hair class pixels and
background class pixels.
[0073] At step 408, a second set of values may be generated by the computing
device 102/1 04 based on the texture of pixels in the two-dimensional ROI image. In
an embodiment, the second set of values corresponds to a second set of LUTs to
store gradient distinctiveness indicator functions for the first class of pixels. In an
embodiment, the computing device 102/1 04 may generate the first set of LUTs and
the second set of LUTs as part of a continuous self-training routine. The computing
device 102/1 04 may utilize Bayesian decision theory along with a set of hair only
distinctive gradient features to determine the region of interest. The computing
device 102/1 04 trains itself to extract these features for every image, based on the
rectangular face-bounding-box information given by the Sony face-tracking (sFace)
system, for example. Notwithstanding, the disclosure may not be so limited, and
other types of face-tracking or facial motion capture systems may be utilized to
provide the face-bounding-box information, without limiting the scope of the
disclosure. The face two-dimensional ROI image is processed by the computing
device 102/104 to identify hair pixels using the self-learn LUTs and the learnt
features.
[0074] At step 4 10, the region of interest is determined by the computing device
102/1 04. At step 4 10a, a first class of pixels and a second class of pixels are
determined based on the first set of values and Bayes rule. At step 4 10b, the
texture first class of pixels and a second class of pixels are analyzed based on the
second set of values. For example, the second set of values corresponds to a
second set of LUTs that store gradient distinctiveness indicator functions for the
first class of pixels. At step 4 12, the method 400 ends.
[0075] With reference to FIGS. 5-1 2D, an example of image processing to
determine the region of interest (ROI) will be described. In this example, the region
of interest is a combination of a head region and a hair region. The computing
device 102/1 04 determines the region of interest using a hair segmentation
algorithm, in accordance with embodiments of the disclosure. The definition of the
interface function for the segmentation algorithm is hairSeg(..). The computing
device 102/1 04 is operable to execute the function hairSeg(..) to perform hair
segmentation to determine the hair region, in accordance with an embodiment of
the disclosure. The input parameters for the interface function hairSeg(..) include:
luminance (L) of the two-dimensional (2D) input image , width (W) of the 2D input
image, height (H) of the 2D input image , x-center of face-bounding box from face
detection (x-coordinate) (face_xc), y-center of face-bounding box from face
detection (y-coordinate) (face_yc), and half width of the face-bounding box
(face_hw).
[0076] The output of the function hairSeg(..) comprises a hair mask image
(pjnask), containing 0s and 1s, which may be a Boolean-type image, and an
optional hair-probability image (p_probMap). Other parameters in the algorithm that
are set default values comprise number of quantization bits (nBits = [6]), a pointer
to a memory for cropped and resized face ROI (facel), width of the facel (szCol =
[75]), and height of the facel (szRow = [75]). The gradient structure for texture
analyses is defined with parameters such as gradient step (step = [1]), y-axis
gradient range (MaxTstepY : [-2 2]), and x-axis gradient range (MaxTstepX : [-2 2]).
Although, certain parameters in the algorithm are set to default values, the values
of these parameters may be changed depending on the application and processing
requirements without limiting the scope of the disclosure.
[0077] The function hairSeg(..) calculates the required size of the texture
(gradient) lookup tables (LUTs) depending on the gradient structure defined.
Luminance LUTs are generated for hair and background classes (hdLumLUT,
bgLumLUT), respectively. Each LUT has a size of 1x2nBlts memory locations with
unsigned integer type. The size of each texture LUT for the gradient structure
disclosed herein may be 24x2nBlts Boolean type. In order to reduce the memory
consumption by the LUTs, the computing device 102/1 04 down quantizes the
cropped face image to the specified nBits parameter value. The down quantization
is performed in real time using the cropAndRszHeadROI(..) function. The down
quantization of the input 2D image reduces the memory requirements for the
histogram based LUTs. The function hairSeg(..) then calls the related subroutines
to perform various sub-tasks of hair segmentation as described with references to
FIGS. 5-1 2D.
[0078] FIG. 5 is a flow chart illustrating a sequence of sub-tasks for hair
segmentation, in accordance with an embodiment of the disclosure. The computing
device 102/1 04 is operable to execute one or more subroutines to perform the
various sub-tasks of hair segmentation to determine the hair region, in accordance
with embodiments of the disclosure. With reference to FIG. 5, there is shown a flow
chart 500 depicting a sequence of sub-routines corresponding to the sub-tasks of
cropping and resizing, generating self-learn lookup tables (LUTs) and distinct
features, and segmenting hair using Bayes rule and distinct gradient functions. The
function hairSeg(..) calls these subroutines to perform various sub-tasks of hair
segmentation.
[0079] At step 502, the computing device 102/1 04 executes the
cropAndRszHeadROI(..) function to generate a two-dimensional face ROI image.
[0080] At step 504, the computing device 102/1 04 executes the
makeTheLUTs(..) function to generate self-learn lookup tables (LUTs) and distinct
features.
[0081] At step, 506, the computing device 102/1 04 executes the
BayesRuleSeg(..) function to segment the hair using Bayes rule and distinct
gradient functions to determine the head region.
[0082] FIGS. 6A-6B are diagrams illustrating screenshots of a two-dimensional
input image and a down-quantized image, in accordance with an embodiment of
the disclosure. With reference to FIG. 6A, there is shown a two-dimensional input
image having a resolution of 8 bits per pixel, for example. With reference to FIG.
6B, there is shown a down-quantized image having a resolution of 6 bits per pixel,
for example. The input to the cropAndRszHeadROI(..) function comprises the twodimensional
input image and its face-bounding-box information. The output of the
cropAndRszHeadROI(..) function comprises a broader face region of interest (ROI)
by cropping, resizing and down quantizing the two-dimensional input image to prespecified
dimensions. The cropping, resizing and down-quantization operations
reduce the memory requirement and computational cost for real-time processing of
the two-dimensional images.
[0083] FIGS. 7A-7B are diagrams illustrating screenshots of a two-dimensional
input image and a cropped, resized and down-quantized image, in accordance with
an embodiment of the disclosure. With reference to FIG. 7A, there is shown the
two-dimensional input image superimposed with the face bounding box. With
reference to FIG. 7B, there is shown a cropped, resized and down-quantized twodimensional
face ROI image. The subsequent processing described with reference
to FIGS. 8-1 2D is performed on the two-dimensional face ROI image illustrated in
FIG. 7B.
[0084] FIG. 8 is a diagram illustrating a screenshot of a two-dimensional ROI
image showing predicted locations for different classes of pixels, in accordance with
an embodiment of the disclosure. With reference to FIG. 8, there is shown a twodimensional
image with predicted class sample areas. The predicted class sample
areas include a head/hair exemplar area, a background exemplar area and an
unspecified area. As seen in FIG. 8, the hair class may comprise a mixture of hair,
face and background classes. Similarly, the background class may include hair
pixels in certain face examples.
[0085] The two-dimensional ROI image with predicted class sample areas of
FIG. 8 constitutes the input to the makeThel_UTs(..) function. The makeThel_UTs(..)
function generates two sets of lookup tables (LUTs), namely, histogram-based
class-conditional probability density functions, also referred to as luminance-based
LUTs, and gradient-based distinct indicator functions (GraDilndicator), also referred
to as texture LUTs. The luminance-based LUTs are formed based on histogram
counts of each luminance value for the background and the hair predicted areas
illustrated in FIG. 8 .
[0086] FIGS. 9A-9B are diagrams illustrating sample histogram plots for hair and
background areas of the two-dimensional ROI image, respectively, in accordance
with an embodiment of the disclosure. With reference to FIG. 9A, there is shown an
area normalized sample histogram plot for hair area of the two-dimensional ROI
image. With reference to FIG. 9B, there is shown an area normalized sample
histogram plot for background area of the two-dimensional ROI image.
[0087] In order to make the LUTs independent of the sample size, the LUTs are
area normalized such that the LUTs, that is, the area under the histogram plots sum
to a same predetermined number (N), independent of the area size over which the
data is collected. For example,
haLUT{k) = bgLUT{k) =N, where nBits = 6, and N = 4096.
k=0 k=0
[0088] The LUT graphs illustrated in FIGS 9A-9B represent class-conditional
probability distribution functions as p(LumValue|Class) in the Bayes rule image
segmentation algorithm (BayesRuleSeg(..) function). In an embodiment, the
luminance-based LUTs are also generated for a skin class, where additional
classification of pixels is required for image segmentation.
[0089] The texture lookup tables (LUTs) are Boolean type. In an embodiment,
the texture LUTs are generated for the hair class only. FIG. 10 is a diagram
illustrating gradient-based distinctiveness functions for hair class as compared to
background class, in accordance with an embodiment of the disclosure. In this
example, it is assumed that the background pixels in the hair area may have similar
texture properties with the selected background area. At the same time, there may
be certain hair texture features that are unique against the background texture
(gradient) features. Based on these criteria, if a hair texture feature is unique, a
value of 1 is assigned to a corresponding index in the LUT, otherwise a value of 0
is assigned, as shown in FIG. 10.
[0090] The gradient based distinctiveness indicator functions are derived based
on the uniqueness of a gradient. The algorithm first (i) defines a gradient direction
with respect to each hair area pixel in the two-dimensional image. For example, the
first gradient direction may be represented by D = 1 and Aj=0. For all the hair area
pixels, the algorithm (ii) initially sets the hair gradient indicators to 1 for each
gradient index (g) as follows:
g = |halmg(i,j)- halmg(i+Ai,j+ Aj)|
haGraDilndicatorl (g) = 1.
[0091] The algorithm then (iii) evaluates the uniqueness of the hair gradient
indicator against the background class based on the texture LUTs. Similarly, for all
the background area pixels, the algorithm resets the hair gradient indicators to 0 if
there is a same gradient in the background class as follows:
g = |bglmg(i,j)- bglimg(i+Ai,j+ Aj)|
haGraDilndicatorl (g) = 0, where "halmg" and "bglmg" represent the hair and
background exemplar areas respectively in FIG. 8 .
[0092] After the evaluation of the hair gradient indicators, (iv) the
haGraDilndicator1 (.) indices that are not reset represent the hair region uniquely.
Additionally, the steps (i) through (iv) are repeated for other gradient directions. In
this example, 24 gradient direction functions are utilized. These gradient direction
functions are designed to be Boolean type, and therefore use minimal amount of
memory.
[0093] The computing device 102/104 determines the region of interest based
on the BayesRuleSeg(..) function. The BayesRuleSeg(..) function utilizes the classconditional
probability functions and the distinct gradient indicator functions to
segment hair regions. The segmentation is based on two parts, namely, Bayes
decision rule and distinctive gradient functions.
[0094] The Bayes decision rule decides whether a pixel in the hair region has to
be assigned to the hair class based on the following expression:
Decide hair if
p(lum Ihair) p(hair) > p(lum \ bg) p(bg);
where prior probability of hair p(hair)= prior probability of face p(face) = 1, and prior
probability of background p(bg) = 3 (or alternately ¼ and ¾). The prior probability of
background is set three times higher than the prior probability of hair. Additionally,
the Bayes decision rule may also include the following expression to distinguish
between the hair regions and the skin regions in the same two-dimensional image:
Decide hair if
p(lum Ihair) p(hair) > p(lum \ bg) p(bg) AND > p(lum \ skin) p(skin);
where p(face)= p(skin).
[0095] In the distinctive gradient analysis, if any one of the gradient
distinctiveness functions indicates that a pixel location is hair, then that pixel is
assigned to the hair class as follows:
Check for all k-directions: if one or more haGraDilndicator-k (.) = 1, then assign the
pixel to the hair class.
Similarly, each pixel in the hair class region of the two-dimensional ROI
image is assigned to the hair class when either the Bayes decision rule or the
distinctive gradient analysis determines that the pixels belongs to the hair class. In
an embodiment, each pixel in the hair class region of the two-dimensional ROI
image is assigned to the hair class when both the Bayes decision rule and the
distinctive gradient analysis determine that the pixels belongs to the hair class.
[0096] FIG. 11 is a diagram illustrating a screenshot of a Boolean-type image
showing the region of interest and a background region, in accordance with an
embodiment of the disclosure. With reference to FIG. 11, there is shown a Booleantype
image which is a sample result of the BayesRuleSeg(..) function based on
Bayes rule decision and gradient distinctiveness functions.
[0097] FIG. 12 is a diagram illustrating a method for post processing of the
determined region of interest, in accordance with an embodiment of the disclosure.
FIG. 12 will be described in conjunction with elements of FIG. 1 and FIG. 2 . The
method 1200 may be implemented in a computing device 102/1 04 comprising one
or more processors 202.
[0098] The method 1200 starts at 1202. At step 1204, the computing device
102/1 04 may subject the region of interest to image smoothing by filling in the gaps
within the region of interest. FIG. 13A is a diagram illustrating a screenshot of a
smoothed image of the region of interest, in accordance with an embodiment of the
disclosure. The outliers, or disconnected patches, around the smoothed region of
interest are removed by the computing device 102/1 04. FIG. 13B is a diagram
illustrating screenshots of a smoothed image of the region of interest before and
after eliminating outlier patches, in accordance with an embodiment of the
disclosure. The computing device 102/1 04 executes a function
enforceBorderConsist(..) to perform image smoothing and eliminate the outliers.
[0099] At step 1206, the computing device 102/104 may detect the outer
contour of the region of interest to generate a region-of-interest or head mask
image. In an embodiment, the computing device 102/104 smoothes the outer
contour by a simple moving average operation. The computing device 102/104
creates the hair mask image by filling the area inside the contour with the
confidence that the contour represents the head (hair plus face) region. The head
mask image is used to assign depth to the head region during 2D to 3D conversion.
The computing device 102/104 executes a function hairBorderDetect(..) to detect
the outer contour and generate the head mask image. FIG. 13C is a diagram
illustrating a screenshot of the region of interest mask, in accordance with an
embodiment of the disclosure.
[00100] At step 1208, the computing device 102/104 may restore the original
scale of the hair mask image to the scale of the two-dimensional input image. The
computing device 102/1 04 executes a function imprintTheFaceMask(..) to restore
the scaling of the resulting head mask image. The scale of the head mask image is
restored to establish correct correspondence between the hair (or head) mask and
the two-dimensional input image to imprint the head mask on the two-dimensional
input image. FIG. 13D is a diagram illustrating a screenshot of the mask image
superimposed on the two-dimensional input image, in accordance with an
embodiment of the disclosure. At step 121 0, the method 1200 ends.
[00101] In accordance with another embodiment of the disclosure, a method
and system for image processing may comprise a computing device 102/104
communicatively coupled to a media management device 106.
[00102] The computing device 102/1 04 may comprise one or more
processors, for example, processor 202 (FIG. 2) that may be operable to determine
luminance values of multiple pixels in a subset of a frame of a two-dimensional
image. The processor 202 may be operable to determine texture values of the
pixels in the subset of the frame based on distinctive gradients of the pixels. The
processor 202 may be further operable to identify a subject region and a
background region in the frame of the two-dimensional image based on the
determined luminance values and the determined texture values of the pixels.
[00103] In an embodiment, the processor 202 may be further operable to
generate a first set of values and a second set of values from the first twodimensional
image. The processor 202 may be operable to determine a region of
interest in the first two-dimensional image. In another embodiment, the processor
202 may be operable to determine a region of interest in a second two-dimensional
image. The processor 202 may be operable to determine a first class of pixels and
a second class of pixels in the second two-dimensional image based on the first set
of values. The processor 202 may be operable to analyze texture of the first class
of pixels and the second class of pixels in the second two-dimensional image
based on the second set of values.
[00104] The system may further comprise a memory (for example, database
memory or memory 204) operable to store a first set of lookup tables (LUTs)
corresponding to the first set of values and a second set of LUTs corresponding to
the second set of values. The first set of LUTs store histogram counts of luminance
values of the first class of pixels and the second class of pixels of the first twodimensional
image. The second set of LUTs store the gradient distinctiveness
indicator functions for the first class of pixels of the first two-dimensional image.
[00105] Other embodiments of the disclosure may provide a non-transitory
computer readable medium and/or storage medium, and/or a non-transitory
machine readable medium and/or storage medium, having stored thereon, a
machine code and/or a computer program having at least one code section
executable by a machine and/or a computer, thereby causing the machine and/or
computer to perform the steps comprising determining a subject region and a
background region in a two-dimensional image. Luminance values of multiple pixels
in a subset of a frame of the two-dimensional image are determined. Texture values
of the multiple pixels in the subset of the frame are determined. The subject region
and the background region in the frame of the two-dimensional image are identified
based on the determined luminance values and the determined texture values of
the multiple pixels
[00106] In an embodiment, the computing device 102/104 may be further
operable to perform automatic depth assignment to human faces in images for 2Dto-
3D image/video conversion. In another embodiment, the results of the
determination of the subject region and the background region may be utilized by
external systems to perform automatic depth assignment to human faces for 2D-to-
3D image/video conversion. In addition, the determination of the subject region and
the background region in accordance with embodiments of the disclosure may have
applications in face recognition, age/gender recognition, digital hair styling, and
human machine interface applications, such as gesture recognition and head pose
estimation.
[00107] Accordingly, the present disclosure may be realized in hardware, or a
combination of hardware and software. The present disclosure may be realized in
a centralized fashion in at least one computer system or in a distributed fashion
where different elements may be spread across several interconnected computer
systems. Any kind of computer system or other apparatus adapted for carrying out
the methods described herein may be suited. A combination of hardware and
software may be a general-purpose computer system with a computer program
that, when being loaded and executed, may control the computer system such that
it carries out the methods described herein. The present disclosure may be realized
in hardware that comprises a portion of an integrated circuit that also performs
other functions.
[00108] The present disclosure may also be embedded in a computer
program product, which comprises all the features enabling the implementation of
the methods described herein, and which when loaded in a computer system is
able to carry out these methods. Computer program in the present context means
any expression, in any language, code or notation, of a set of instructions intended
to cause a system having an information processing capability to perform a
particular function either directly or after either or both of the following: a)
conversion to another language, code or notation; b) reproduction in a different
material form.
[00109] While the present disclosure has been described with reference to
certain embodiments, it will be understood by those skilled in the art that various
changes may be made and equivalents may be substituted without departing from
the scope of the present disclosure. In addition, many modifications may be made
to adapt a particular situation or material to the teachings of the present disclosure
without departing from its scope. Therefore, it is intended that the present
disclosure not be limited to particular embodiments disclosed, but that the present
disclosure will include all embodiments falling within the scope of the appended
claims.

CLAIMS
What is claimed is:
1. A method for image processing, said method comprising:
in a computing device:
determining luminance values of a plurality of pixels in a subset of a
frame of a two-dimensional image;
determining texture values of said plurality of pixels in said subset of
said frame; and
identifying a subject region and a background region in said frame of
said two-dimensional image based on said determined luminance values
and said determined texture values of said plurality of pixels.
2 . The method of claim 1, comprising storing said luminance values and said
texture values of said plurality of pixels in a first set of lookup tables and a
second set of lookup tables respectively.
3 . The method of claim 1, wherein said identification of said subject region and
said background region in said frame of said two-dimensional image comprises
classifying said plurality of pixels into a first class and/or a second class based
on a comparison of said luminance values and said texture values of said
plurality of pixels with a first set of lookup tables and a second set of lookup
tables respectively.
4 . The method of claim 3, wherein said first class comprises a head class and said
second class comprises a background class, wherein said head class
comprises a face class and a hair class.
5 . The method of claim 1, comprising converting said two-dimensional image to a
three-dimensional image based on said identification of said subject region and
said background region in said two-dimensional image.
6 . A method for image processing, said method comprising:
in a computing device:
generating a first set of values based on luminance of a first class of
pixels and a second class of pixels in a first two-dimensional image; and
generating a second set of values based on texture of said first class
of pixels and said second class of pixels in said first two-dimensional image,
wherein said generated first set of values and said generated second set of
values are used to classify a subject pixel into one of a first class of pixels
and a second class of pixels based on a comparison of a luminance value of
said subject pixel and a texture value of said subject pixel with said
generated first set of values and said generated second set of values
respectively.
7 . The method of claim 6, wherein said subject pixel belongs to said first two
dimensional image.
8 . The method of claim 6, wherein said subject pixel belongs to a second twodimensional
image different from said first two-dimensional image.
9 . The method of claim 6, wherein said first set of values corresponds to a first set
of lookup tables that store histogram counts of luminance values of said first
class of pixels and said second class of pixels in said first two-dimensional
image.
10 .The method of claim 6, wherein said second set of values corresponds to a
second set of lookup tables that store gradient distinctiveness indicator
functions for said first class of pixels.
11.A method for image processing, said method comprising:
in a computing device:
determining a region of interest in a two-dimensional image,
comprising:
determining a first class of pixels and a second class of pixels
in said two dimensional image based on a first set of predetermined
values corresponding to a luminance of said two-dimensional image;
and
analyzing texture of said first class of pixels and said second
class of pixels based on a second set of predetermined values
corresponding to said texture of said two-dimensional image; and
converting said two-dimensional image to a three-dimensional image
based on said determination of said region of interest in said two-dimensional
image.
12 .The method of claim 11, further comprising generating said two-dimensional
image from a two-dimensional input image, wherein said two-dimensional image
has a lower dimension compared to said two-dimensional input image.
13 .The method of claim 12, wherein said generation of said two-dimensional image
from said two-dimensional input image comprises cropping, resizing and/or
down-quantizing said two-dimensional input image.
14. The method of claim 11, wherein said determination of said first class of pixels
and said second class of pixels in said two-dimensional image comprises:
predicting a first region and a second region in said two-dimensional image
corresponding to said first class of pixels and said second class of pixels; and
computing class-conditional probability functions based on luminance of said
first and second class of pixels and said first set of predetermined values, wherein
said class-conditional probability functions are computed using Bayes rule.
15 .The method of claim 14, wherein said predicted first region and said predicted
second region comprise a head region and a background region respectively.
16 .The method of claim 11, wherein said first set of predetermined values
corresponds to a first set of lookup tables to store histogram counts of
luminance values of said first class of pixels and said second class of pixels of
one or more of said two-dimensional image and at least one previous twodimensional
image.
17 .The method of claim 11, wherein said analyses of said texture comprises:
computing gradient distinctiveness indicator functions for said first class of
pixels and said second class of pixels;
determining uniqueness of said first class of pixels as compared to said
second class of pixels based on said computed gradient distinctiveness indicator
functions and said second set of predetermined values.
18 .The method of claim 11, wherein said second set of predetermined values
corresponds to a second set of lookup tables to store said gradient
distinctiveness indicator functions for said first class of pixels of one or more of
said two-dimensional image and at least one previous two-dimensional image.
19 .The method of claim 11, wherein said first class comprises a head class and
said second class comprises a background class, wherein said head class
comprises a face class and a hair class.
20. The method of claim 11, wherein said determined region of interest comprises
one or both: of a hair region and/or a head region.
2 1.The method of claim 11, wherein said determined region of interest is
represented as a Boolean-type image.
22. The method of claim 11, further comprising:
smoothing said determined region of interest;
eliminating outliers around said smoothed region of interest;
detecting an outer contour of said region of interest to generate a region of
interest mask image; and
restoring said region of interest mask image to an original scale of a twodimensional
input image.
23.A system for image processing, said system comprising:
one or more processors in said computing device being operable to:
determine luminance values of a plurality of pixels in a subset of a
frame of a two-dimensional image;
determine texture values of said plurality of pixels in said subset of
said frame based on distinctive gradients of said plurality of pixels; and
identify a subject region and a background region in said frame of
said two-dimensional image based on said determined luminance values
and said determined texture values of said plurality of pixels.
24. The system of claim 23, further comprising a memory operable to store:
a first set of lookup tables corresponding to said determined luminance
values; and
a second set of lookup tables corresponding to said determined texture
values.
25. The system of claim 23, wherein said one or more processors are operable to
identify said subject region and said background region in said frame of said
two-dimensional image by classifying said plurality of pixels into a first class
and/or a second class based on a comparison of said luminance values and
said texture values of said plurality of pixels with a first set of lookup tables and
a second set of lookup tables respectively.
26. The system of claim 25, wherein said first class comprises a head class and
said second class comprises a background class, wherein said head class
comprises a face class and a hair class.
27. The system of claim 23, wherein said one or more processors are operable to
convert said two-dimensional image to a three-dimensional image based on
said identification of said subject region and said background region in said twodimensional
image.
28.A non-transitory computer-readable storage medium having stored thereon, a
computer program having at least one code section for image processing, the at
least one code section being executable by a computer for causing said
computer to perform steps comprising:
determining luminance values of a plurality of pixels in a subset of a
frame of a two-dimensional image;
determining texture values of said plurality of pixels in said subset of
said frame; and
identifying a subject region and a background region in said frame of
said two-dimensional image based on said determined luminance values
and said determined texture values of said plurality of pixels.