Claims:1. A method for stamp image detection using unsupervised feature learning techniques; said method comprising processor implemented steps of:
receiving, a training data comprising a plurality of stamp images, using an image capturing device (200);
sampling, a plurality of square patches of same size from one or more of the plurality of stamp images, using a sampling module (210);
generating, a plurality of whitened patches by performing zero-phase component analysis (ZCA) whitening on the plurality of square patches, using a image processing module (212);
obtaining, a plurality of dictionary atoms by performing K-means clustering on the plurality of whitened patches, using the image processing module (212);
calculating, a response for each of the plurality of dictionary atoms and ranking the plurality of dictionary atoms based on the calculated responses, using a ranking module (214); and
detecting one or more stamp image on a first test image using the ranked dictionary atoms by a stamp detection and verification module (218).

2. The method according to claim 1 comprising verification of at least one stamp image on a second test image wherein verification comprises:
extracting, at least one feature vector from a test image, using the ranked dictionary atoms, by a feature extraction module (216); and
verifying, the second test image using the at least one feature vector, by the stamp detection and verification module (218).

3. The method according to claim 1 wherein ranking comprises:
selecting, at least one stamp image form the plurality of stamp images;
selecting overlapping square patches of same size from the at least one stamp image to generate a patch set;
projecting the patch set on the plurality of dictionary atoms;
implementing thresholding, using rectified liner unit(ReLu) to generate a response for each of the plurality of dictionary atoms corresponding to each of the plurality of overlapping patch;
selecting the maximum response out of the responses for each of the plurality of dictionary atoms corresponding to each of the plurality of overlapping patch; and
ranking the plurality of dictionary atoms based on the calculated response.

4. The method according to claim 2 wherein extracting of the at least one feature vector comprises:
convolving, the test image, with the obtained dictionary atoms to generate a convolved test image;
encoding, the convolved test image using 1-of-K, max-assignment to select maximum of K values;
performing 4 x 4 -quadrant max pooling on each feature map of the convolved test image; and
concatenating, each of the feature maps to extract the at least one feature vector.

5. A system (102) for stamp image detection using unsupervised feature learning techniques; said system (102) comprising an image capture device (200) operatively coupled to the system (102), a processor (202), an interface (204), and memory (206), the system comprising:
an image capture device (200) configured to receive, a training data comprising a plurality of stamp images;
a sampling module (210) configured to sample, a plurality of square patches of same size from one or more of the plurality of stamp images;
an image processing module (212) generate, a plurality of whitened patches by performing zero-phase component analysis (ZCA) whitening on the plurality of square patches;
the image processing module (212) configured to obtain, a plurality of dictionary atoms by performing K-means clustering on the plurality of whitened patches;
a ranking module (214) configured to calculate, a response for each of the plurality of dictionary atoms and ranking the plurality of dictionary atoms in descending order of the calculated responses; and
a stamp detection and verification (218) module configured to detect one or more stamp image on a first test image using the ranked dictionary atoms.
6. The system (102) of claim 1 further configured to verify at least one stamp image on a second test image wherein in order to verify the at least one stamp image on the second test image the system (102) is adapted for
extracting, at least one feature vector from a test image, using the ranked dictionary atoms, by a feature extraction module (216); and
verifying, the second test image using the at least one feature vector, by the stamp detection and verification module (218).

7. The system according to claim 5, wherein the ranking module (214) is further configured to:
select, at least one stamp image form the plurality of stamp images;
select overlapping square patches of same size from the at least one stamp image to generate a patch set;
project the patch set on the plurality of dictionary atoms;
implement, thresholding, using rectified liner unit(ReLu) to generate a response for each of the plurality of dictionary atoms corresponding to each of the plurality of overlapping patch;
select, the maximum response out of the responses for each of the plurality of dictionary atoms corresponding to each of the plurality of overlapping patch; and
rank, the plurality of dictionary atoms based on the maximum response.

8. The system according to claim 6, wherein the feature extraction module (216) is configured to:
convolve, the test image, with the obtained dictionary atoms to generate a convolved test image;
encode, the convolved test image using 1-of-K, max-assignment to select maximum of the K values;
perform 4 x 4 -quadrant max pooling on each feature map of the convolved test image; and
concatenate, each of the feature maps to extract the at least one feature vector.
, Description:
FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:
METHOD AND SYSTEM FOR STAMP IMAGE DETECTION AND VERIFICATION USING UNSUPERVISED FEATURE LEARNING TECHNIQUES

Applicant:
Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th floor,
Nariman point, Mumbai 400021,
Maharashtra, India

The following specification particularly describes the invention and the manner in which it is to be performed.
FIELD OF THE INVENTION

[001] The present application generally relates to machine learning. Particularly, the application provides a method and system for stamp detection and verification.

BACKGROUND OF THE INVENTION

[002] In countries like India, several government, bank, real estate etc. related transactions take place on paper. There is a strong recent initiative to reduce paper based transaction, however digitization of archival data remains a big challenge for achieving this goal. Detecting and verifying stamps in documents is an important problem since stamps can be indicators of authenticity.

[003] Prior art illustrates to an unsupervised feature learning approach for learning an appropriate representation for stamp shapes. Recently, Adam Coates, Andrew Ng and Honglak Lee, G in An Analysis of Single-Layer Networks in Unsupervised Feature Learning, ((JMLR W-CP), 15:215-223, 2011) state that the single layer of convolution filters learned with an unsupervised dictionary learning method such as K-means clustering performs well on object recognition.

[004] However, obtaining high recognition rates based on a relatively low number of dictionary atoms is an area not explored by the prior art literature. Selecting the low number of dictionary atoms out of the dictionary atoms is a relatively un-explored area, however the same may be used for efficiently detecting and verifying of stamp images on a document.

[005] An efficient method for automatically selecting dictionary atoms for efficient detection and verification of stamp images is therefore a technical challenge in the current state of the art.
SUMMARY OF THE INVENTION

[006] Before the present methods, systems, and hardware enablement are described, it is to be understood that this invention is not limited to the particular systems, and methodologies described, as there can be multiple possible embodiments of the present invention which are not expressly illustrated in the present disclosure. It is also to be understood that the terminology used in the description is for the purpose of describing the particular versions or embodiments only, and is not intended to limit the scope of the present invention which will be limited only by the appended claims.

[007] The present application provides a method and system for stamp image detection and verification using unsupervised feature learning.

[008] In one aspect the present application provides a method for stamp image detection using unsupervised feature learning techniques; said method comprising processor implemented steps of receiving, a training data comprising a plurality of stamp images, using an image capturing device (200). The method further comprises sampling, a plurality of square patches of same size from one or more of the plurality of stamp images, using a sampling module (210). Further the method comprises, generating, a plurality of whitened patches by performing zero-phase component analysis (ZCA) whitening on the plurality of square patches, using an image processing module (212). Further the disclosed method comprises obtaining, a plurality of dictionary atoms by performing K-means clustering on the plurality of whitened patches, using the image processing module (212). The proposed method further comprises calculating, a response for each of the plurality of dictionary atoms and ranking the plurality of dictionary atoms based on the calculated responses, using a ranking module (214) and detecting one or more stamp image on a first test image using the ranked dictionary atoms by a stamp detection and verification module (218).

[009] Further the disclosed subject matter discloses verification of at least one stamp image on a second test image wherein verification comprises extracting, at least one feature vector from a test image, using the ranked dictionary atoms, by a feature extraction module (216); and verifying, the second test image using the at least one feature vector, by the stamp detection and verification module (218).

[0010] In another aspect the present application provides a system (102) for stamp image detection using unsupervised feature learning techniques; said system (102) comprising an image capture device (200) operatively coupled to the system (102), a processor (202), an interface (204), and memory (206), the system comprises an image capture device (200) configured to receive, a training data comprising a plurality of stamp images, the system further comprises a sampling module (210) configured to sample, a plurality of square patches of same size from one or more of the plurality of stamp images. The system (102) further comprises an image processing module (212) generate, a plurality of whitened patches by performing ZCA whitening on the plurality of square patches. In an embodiment, the image processing module (212) configured to obtain, a plurality of dictionary atoms by performing K-means clustering on the plurality of whitened patches. The system (102) further comprises a ranking module (214) configured to calculate, a response for each of the plurality of dictionary atoms and ranking the plurality of dictionary atoms in descending order of the calculated responses and a stamp detection and verification module (218) module configured to detect one or more stamp image on a first test image using the ranked dictionary atoms.

[0011] Further The system (102) of claim 1 may be configured to verify at least one stamp image on a second test image wherein in order to verify the at least one stamp image on the second test image the system (102) is adapted for extracting, at least one feature vector from a test image, using the ranked dictionary atoms, by a feature extraction module (216), and verifying, the second test image using the at least one feature vector, by the stamp detection and verification module (218).

BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The foregoing summary, as well as the following detailed description of preferred embodiments, are better understood when read in conjunction with the appended drawings. For the purpose of illustrating the invention, there is shown in the drawings exemplary constructions of the invention; however, the invention is not limited to the specific methods and system disclosed. In the drawings:

[0013] Figure 1: shows a network implementation (100) of a system (102) for stamp detection and verification in accordance with an embodiment of the disclosed subject matter;

[0014] Figure 2: shows a block diagram illustrating the system (102) for stamp detection and verification in accordance with an embodiment of the disclosed subject matter;

[0015] Figure 3: shows a flow chart illustrating steps for stamp detection and verification in accordance with an embodiment of the disclosed subject matter;

[0016] Figure 4(a): shows a learned dictionary for K= 64 in accordance with an exemplary embodiment of the disclosed subject matter;

[0017] Figure 4(b): shows a ranked dictionary atoms for K=64 where the selected portion is the subset of ranked dictionary atoms picked for detection and verification of stamp images in accordance with an exemplary embodiment of the disclosed subject matter;

[0018] Figure 5: shows a flowchart illustrating steps for ranking dictionary atoms for detecting and verifying stamp images in accordance with an embodiment of the disclosed subject matter;

[0019] Figure 6: shows a flowchart illustrating steps for feature extraction for detecting and verifying stamp images in accordance with embodiment of the disclosed subject matter.

DETAILED DESCRIPTION OF THE INVENTION

[0020] Some embodiments of this invention, illustrating all its features, will now be discussed in detail.

[0021] The words "comprising," "having," "containing," and "including," and other forms thereof, are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items.

[0022] It must also be noted that as used herein and in the appended claims, the singular forms "a," "an," and "the" include plural references unless the context clearly dictates otherwise. Although any systems and methods similar or equivalent to those described herein can be used in the practice or testing of embodiments of the present invention, the preferred, systems and methods are now described.

[0023] The disclosed embodiments are merely exemplary of the invention, which may be embodied in various forms.

[0024] The elements illustrated in the Figures inter-operate as explained in more detail below. Before setting forth the detailed explanation, however, it is noted that all of the discussion below, regardless of the particular implementation being described, is exemplary in nature, rather than limiting. For example, although selected aspects, features, or components of the implementations are depicted as being stored in memories, all or part of the systems and methods consistent with the attrition warning system and method may be stored on, distributed across, or read from other machine-readable media.

[0025] The techniques described above may be implemented in one or more computer programs executing on (or executable by) a programmable computer including any combination of any number of the following: a processor, a storage medium readable and/or writable by the processor (including, for example, volatile and non-volatile memory and/or storage elements), plurality of input units, and plurality of output devices. Program code may be applied to input entered using any of the plurality of input units to perform the functions described and to generate an output displayed upon any of the plurality of output devices.

[0026] Each computer program within the scope of the claims below may be implemented in any programming language, such as assembly language, machine language, a high-level procedural programming language, or an object-oriented programming language. The programming language may, for example, be a compiled or interpreted programming language. Each such computer program may be implemented in a computer program product tangibly embodied in a machine-readable storage device for execution by a computer processor.

[0027] Method steps of the invention may be performed by one or more computer processors executing a program tangibly embodied on a computer-readable medium to perform functions of the invention by operating on input and generating output. Suitable processors include, by way of example, both general and special purpose microprocessors. Generally, the processor receives (reads) instructions and data from a memory (such as a read-only memory and/or a random access memory) and writes (stores) instructions and data to the memory. Storage devices suitable for tangibly embodying computer program instructions and data include, for example, all forms of non-volatile memory, such as semiconductor memory devices, including EPROM, EEPROM, and flash memory devices; magnetic disks such as internal hard disks and removable disks; magneto-optical disks; and CD-ROMs. Any of the foregoing may be supplemented by, or incorporated in, specially-designed ASICs (application-specific integrated circuits) or FPGAs (Field-Programmable Gate Arrays). A computer can generally also receive (read) programs and data from, and write (store) programs and data to, a non-transitory computer-readable storage medium such as an internal disk (not shown) or a removable disk.

[0028] Any data disclosed herein may be implemented, for example, in one or more data structures tangibly stored on a non-transitory computer-readable medium. Embodiments of the invention may store such data in such data structure(s) and read such data from such data structure(s).

[0029] The present application provides a computer implemented method and system for unsupervised word image clustering.

[0030] Referring to Fig. 1, a network implementation 100 of a system 102 for stamp detection using unsupervised feature learning techniques, in accordance with an embodiment of the present subject matter. Although the present subject matter is explained considering that the system 102 is implemented on a server, it may be understood that the system 102 may also be implemented in a variety of computing systems, such as a laptop computer, a desktop computer, a notebook, a workstation, a mainframe computer, a server, a network server, and the like. In one implementation, the system 102 may be implemented in a cloud-based environment. It will be understood that the system 102 may be accessed by multiple users through one or more user devices 104-1, 104-2…104-N, collectively referred to as user devices 104 hereinafter, or applications residing on the user devices 104. Examples of the user devices 104 may include, but are not limited to, a portable computer, a personal digital assistant, a handheld device, and a workstation. The user devices 104 are communicatively coupled to the system 102 through a network 106.

[0031] In one implementation, the network 106 may be a wireless network, a wired network or a combination thereof. The network 106 can be implemented as one of the different types of networks, such as intranet, local area network (LAN), wide area network (WAN), the internet, and the like. The network 106 may either be a dedicated network or a shared network. The shared network represents an association of the different types of networks that use a variety of protocols, for example, Hypertext Transfer Protocol (HTTP), Transmission Control Protocol/Internet Protocol (TCP/IP), Wireless Application Protocol (WAP), and the like, to communicate with one another. Further the network 106 may include a variety of network devices, including routers, bridges, servers, computing devices, storage devices, and the like.

[0032] In one embodiment of the invention, referring to Fig. 2 block diagram illustrating a system (102) for stamp detection using unsupervised feature learning techniques is disclosed. The system (102) comprises at least one image capture device (200) configured to receive, a training data. In an embodiment, the training data comprises a plurality of stamp images. The system (102) further comprises a sampling module (210) which is configured to sample, a plurality of square patches of same size from one or more of the plurality of stamp images. In an embodiment the sampled patches may be of size m x m. The system further comprises an image processing module (212) configured to generate, a plurality of whitened patches by performing zero-phase component analysis (ZCA) whitening on the plurality of square patches. In another embodiment the image processing module may be further configured to obtain, a plurality of dictionary atoms by performing K-means clustering on the plurality of whitened patches.

[0033] The system (102) further comprises a ranking module (214) configured to calculate, a response for each of the plurality of dictionary atoms and ranking the plurality of dictionary atoms in descending order of the calculated responses. In one embodiment the ranking module (214) is configured to select, at least one stamp image form the plurality of stamp images. The ranking module (214) select overlapping square patches of same size from the at least one stamp image to generate a patch set and project the patch set on the plurality of dictionary atoms. The ranking module is further configured to implement, thresholding, using rectified liner unit (ReLu) to generate a response for each of the plurality of dictionary atoms corresponding to each of the plurality of overlapping patch and selecting, the maximum response out of the responses for each of the plurality of dictionary atoms corresponding to each of the plurality of overlapping patch. In another embodiment the ranking module (214) is further configured to rank, the plurality of dictionary atoms based on the maximum response.

[0034] In an exemplary embodiment the ranking module (214) is configured to randomly select a stamp image from the training data. From the training data, overlapping patches of size m x m are obtained from all pixel locations (i.e. stride is set to 1). In an embodiment Y denotes the patch set. We project Y on the obtained K atoms and perform thresholding using a Rectified Linear unit (ReLu) as follows:
…… (1)

[0035] Referring to equation (1), Rij denotes the response of jth atom for ith patch and n denotes the number of patches in Y. In an embodiment yic denotes the intensity value at the center of the patch. Since stamps are on a lighter background, post multiplication by (1 – yic) assigns more weight to the patch response if it contains a part of stamp. The above operation is equivalent to convolving K filters with the training image, performing rectification on the result and pixel-wise multiplying by an inverted input image. Further response for a dictionary atom is calculated as the maximum of an overall response by implementing equation (2).

…………. (2)

[0036] Referring to equation (2), Sj denotes the maximum response attained by jth atom. In an embodiment of the disclosed subject matter the atoms are in the descending order of their responses. In one embodiment top “v” atoms are selected to be used for stamp detection and verification such that “v” is based on a predefined threshold value of responses.

[0037] The system further comprises a stamp detection and verification module (218) configured to detect one or more stamp images on a first test image using the ranked dictionary atoms by applying filters on the first test image and detecting stamp image.

[0038] In another embodiment the system (102) may further be configured to verify the presence of a stamp on a second test image wherein the second test image may or may not have a stamp image present on the second test image.

[0039] In order to verify the stamp image on the second test image the system (102) comprises a feature extraction module (216) configured to extract at least one feature vector from a test image, using the ranked dictionary atoms. In an embodiment the feature extraction module (216) is configured such that the feature extraction module (216) convolves, the test image, with the obtained dictionary atoms, further the feature extraction module encodes, the convolved test image using 1-of-K, max-assignment to select maximum of the K values. The feature extraction module (216) then performs a 4 x 4 -quadrant max pooling on each feature map of the convolved test image; and finally concatenates, each of the feature maps to extract the at least one feature vector. In an embodiment of the subject matter disclosed herein the 1-of-K, max-assignment is applied as per equation (3).
……………. (3)

[0040] Referring to Fig. 4(a) an exemplary embodiment of a learned dictionary i.e. the result of the K means clustering, is illustrated wherein the value of K is 64. Further Fig 4(b) illustrates a ranked dictionary atoms for K = 64 such that the marked area is the selected subset of the ranked dictionary atoms which may be used for stamp detection and verification.

[0041] Further referring to Fig. 2 the system (102) also comprises a stamp detection and verification module (218) configured to at least one stamp on the test image using the at least one feature vector. In an embodiment the stamp detection and verification module (218) may be configured to verify at least one stamp on the test image.

[0042] Referring now to Fig. 3 a flow chart illustrating steps stamp detection using unsupervised feature learning techniques in accordance with an embodiment of the disclosed subject matter is shown. The process starts at step 302, a training data comprising a plurality of stamp images is received using an image capture device (200) including a camera, scanner and the like. At the step 304, a plurality of square patches of same size are sampled from one or more of the plurality of stamp images. At the step 306 a plurality of whitened patches are generated by performing ZCA whitening on the plurality of square patches.

[0043] At the step 308 a plurality of dictionary atoms are obtained by performing K-means clustering on the plurality of whitened patches. The plurality of dictionary atoms are ranked such that ranking comprises calculating a response for each of the plurality of dictionary atoms and ranking the dictionary atoms based on the calculated response at shown at the step 310.

[0044] In one embodiment one or more stamp images may be detected on a first test image using the ranked dictionary atoms as shown at step 312.

[0045] The proposed method may further be used for verification of a stamp image. In an embodiment, at least one stamp image may be verified for a second test image wherein the second test image may or may not have a stamp image on the second test image. Steps for verification, in accordance with an embodiment, are as follows: at the step 314 at least one feature vector is extracted for a test image such that the at least one feature vector is based on the ranked dictionary atoms. Finally at the step 316 at least one stamp image is verified on the test image using the at least one feature vector.

[0046] Referring now to Fig. 5 a flowchart illustrating the method for ranking of the plurality of dictionary atoms in accordance with the disclosed subject matter is illustrated. At the step 502 at least one stamp image is selected, form the plurality of stamp images. Step 504 illustrates selecting overlapping square patches of same size from the at least one stamp image to generate a patch set. Ranking further comprises as illustrated at the step 506, projecting the patch set on the plurality of dictionary atoms. At the step 508 thresholding, is implemented, using rectified liner unit (ReLu), to generate a response for each of the plurality of dictionary atoms corresponding to each of the plurality of overlapping patch. At the step 510 the maximum response out of the responses for each of the plurality of dictionary atoms corresponding to each of the plurality of overlapping patch is selected. Finally at the step 512 the plurality of dictionary atoms based on the value of response.

[0047] Referring now to Fig. 6 a flowchart illustrating the method for feature extraction in accordance with the disclosed subject matter is illustrated. At the step 602 a test image is convolved with the obtained dictionary atoms to generate a convolved test image. In an embodiment the convolved test image comprises a plurality of patches such that each of the plurality of patches may comprise a feature map. The convolved test image is encoded using 1-of-K, max-assignment to select maximum of K values and performing 4 x 4 -quadrant max pooling on each feature map of the convolved test image as illustrated at the step 604. Finally at the step 606 each of the feature maps of the convolved test image is concatenated to extract the at least one feature vector.

[0048] The following experimental data is added to explain the method disclosed herein and is not intended to limit the scope of the present application. The experimental data may only be used for demonstrate the results of the method disclosed in the present application.

[0049] In the first case for stamp verification, given a test image, our aim is to classify it as a stamp or non-stamp. For obtaining the dataset for non-stamp images, the fact that stamps in our documents always lie in the lower half side is utilized. Patches are sampled from the upper half only. The non-stamp set mainly consists of text regions, background regions or document borders. The training data used consist of 882 stamp and 957 non-stamp images. Prior to feature extraction, all the images are converted to grayscale, resized to a fixed dimension and normalized in the range 0 to 1. A patch size of 16 x 16 is used for the experiment. The feature set is randomly divided in 70%-30% for training and testing respectively. A binary linear SVM classifier is trained on training features and compute classification accuracy on the test set.

[0050] For comparison, the classification was performed with following settings: subset of ranked dictionary atoms (v = 21), versus use all dictionary atoms (v = 64). 64 Gabor filters (8 scale and 8 orientations), 64 Random Filters (RF). Table 1 shows our classification results. A relatively smaller set (approx. 1/3 rd) of ranked dictionary atoms produced a superior performance as compared to the full set (with less testing time). Testing time reported here is with MATLAB implementation. Also the proposed approach significantly outperformed off-the-shelf shape descriptor such as Gabor filters and a single layer of random filter based recognition.

Table 1

[0051] In the second case for stamp detection used for locating a stamp from the images. Firstly top v filters with the input image were convolved and rectification was performed as per equation (1). An average of the responses from the filters was computed and it was observed that, a relatively high response was received at the stamp locations and a low response at non-stamp locations. Using a moving window sum method, a region of maximum response is located. Bounding box of the stamp is then decided by local threshold based heuristic method. Stamp detection performance is measured as an average Intersection over Union (IoU) overlap between the box markings obtained from a proprietary crowd-sourcing experiment and ones which are estimated algorithmically. An average IoU overlap of 74.81% was detected which underlines efficiency of the proposed method.