NON-TEMPLATE-BASED METHOD FOR AUTOMATIC SIGNATURE EXTRACTION FROM CHEQUE IMAGES
FIELD OF THE INVENTION
The instant invention generally relates to a method of extracting the signatures of the user from a bank cheque. More particularly, the present invention relates to an improved method for dynamically extracting the signatures of a user from a bank cheque without the prior knowledge of the banks' cheque template or the user signature.
BACKGROUND OF THE INVENTION
A typical system for signature extraction makes use of a cheque template to match it with the sample cheque and extract the signature. This is useful in cases when the cheques whose templates are already fed to the system arrive for clearance. If however, a bank receives a cheque, which has a different format and no template saved with them, it becomes tedious. The user then has to send such a cheque for manual verification, as automatic extraction is not possible. This wastes a lot of time and extra labor. It even incurs extra cost to the bank, a delay in services and poor customer satisfaction due to delay in the clearance of the cheque. Poor signature detection and verification techniques also lead to higher fraud risks.
Cheques of different banks have different format of cheques. Even the same bank may have two or more different formats to suit the requirements of different type of users like corporate and individual customers. Also, the cheque formats keep evolving with time. This further makes the problem of automatic cheque signature extraction more difficult and complex. Therefore, a format/template-based approach is not feasible in real-life scenario.

The already existing methods in the field include sytems that extract the signatures based on the template format previously stored in the database. In cases when there is no template available, it becomes impossible to perform the automatic extraction of the signatures and the cheque has to be sent for manual extraction and later verification. Also, the banks keep on updating the format of their cheques and with the system being dependent on a particular saved format for the processing of the cheque, it becomes difficult for the extraction of the signatures for verification and clearance. This leads to wastage of time and labor and also induces unnecessary delays in the clearance of the cheque.
Another main area of concern for the already existing methods is that they are not able to extract the signatures and other user-entered information from the cheques due to the problem pos.ed by varied backgrounds and the overlapping/intermixing of machine printed and handwritten information.
Thus, there is a need for a system that focuses on extraction of the signatures without any information or template available beforehand.
OBJECTIVES AND SUMMARY OF THE INVENTION
The present invention provides a non-template method and apparatus for the extraction of the customer's signature from a bank cheque for offline signature verification without the need of any prior information about the user's signature or requirement of any template of the banks' cheque.
The present invention also provides a system and method for cheque fraud detection using signature validation that involves the dynamic extraction of single or multiple signatures from the cheque image using the concept of Most Probable Signature Region (MPSR).-

Another embodiment of the invention provides a capability of extraction of the customers' signature from the bank even if there are no sample signatures available in the database.
An embodiment of the invention provides a system and method that successfully extracts the signatures from the cheque with varied and colored backgrounds.
A yet another embodiment of the invention provides a system and method that extracts signatures even from cheques on which there is/are stamp(s) touching the signature.
An embodiment of the invention provides a system and method that successfully removes the components and retains the signature even when the signature overlapps the machine printed components.
Another embodiment of the invention provides a method and system that uses MPSR concept to detect and extract the signatures even from portions, which are outside the right bottommost quadrant of the cheque.
To implement the aforementioned embodiments, the present invention provides a template independent method for automatic extraction of noise free signature image from a document, said method comprising the steps of:
- scanning the document to obtain a cheque image;
- identifying and defining a Most Probable Signature Region [MPSR] within the cheque image;
- removing the noise from the image
- detecting the stamp from noise removed image
- removing stamp and other unwanted components in the MPSR
region
- obtaining the required signature image.

Also scan line component approach comprises the steps of:
- starting from the bottom-most point of MPSR and drawing a
horizontal straight line across the MPSR
- connecting all the components lying along this line and counting
them as one
- counting the total number of components in the image;
- removing the line drawn in previous step and restoring all the components to their original state
- drawing a horizontal line a pixel above the last drawn line and repeatedly execute above sequence of steps from second step onwards until the entire MPSR region is covered line by line along Y axis;
- marking the point where sudden, significant drop in the number of components is encountered as the bottom of the stamp;
- marking the point where sudden, significant increase in the number of components is encountered as the top of the stamp
- marking region between the bottom point and top point as stamp region
- removing the stamp and other unwanted components from the MPSR using techniques including overlapping region, height analysis, density analysis and histogram analysis and
- cropping image from the MPSR to get a clean signature image for further processing
The present invention further provides for a system for automatic extraction of signature image from a document comprising:
- input means for receiving user inputs to customize data extraction
from the given document

- acquisition means coupled with the input means for reading - in data from the document
- detection unit for detecting the presence and location of the required field on given document
- memory unit coupled with detection unit for storing data as it is being processed
- extraction unit for extracting required information from the given document and an
- output means coupled with extraction unit for providing the extracted information
BRIEF DESCRIPTION OF DRAWINGS
The proposed method and system in described with reference to the accompanying figures. In the figures, the left most digit(s) of a reference number identifies the figure in which the reference number first appears. The same numbers are used throughout the drawings to reference like features and components.
Figure 1 is a flow diagram illustrating the proposed method for automatic extraction of signature image.
Figure 2 is a flow diagram illustrating an exemplary method of removal of noise from signature image.
Figure 3 illustrates the flow diagram representation of the scan line component approach.

Figure 4 is a flow diagram illustrating an exemplary method of removal of components from the image.
Figure 5 a) illustrates an exemplary signature image with associated noise and unwanted data.
Figure 5 b) illustrates an exemplary signature image after removal of associated noise and unwanted data.
Figure 6 a) illustrates an exemplary MPSR image after removal of noise but with presence of stamp.
Figure 6 b) illustrates an exemplary signature image after detection and extraction of stamp.
Figure 7 a) illustrates an exemplary portion of a cheque when the signature overlaps the MICR.
Figure 7 b) illustrates an exemplary signature image after removal of MICR and stamp region.
Figure 8 a) illustrates an exemplary signature image with associated stamp and noise as unwanted data.
Figure 8 b) illustrates an exemplary signature image with noise removed.
Figure 8 c) illustrates an exemplary signature image with noise and stamp both removed.
Figure 9 illustrates an exemplary system for automatic extraction of noise free image from a document.

DETAILED DESCRIPTION OF DRAWINGS
A method for efficient means of automatic extraction of signature image for signature verification is described.
The description of the method shown herein below is intended only for illustration and disclosure of an operative embodiment and not to show all of the various forms or modifications in which this invention might be embodied or operated, since the same may be modified in various particulars or relations without departing from the spirit or scope of the claimed invention.
The present invention describes a non-template based method and system that offers a solution for automatic and efficient extraction of signature image from various instruments and paper-based documents such as bank cheques, drafts, orders etc.
The instant invention aims for a dynamic method of signature extraction from a bank cheque without using a template of the banks' cheque or a sample of the user's signature.
The method described not only extracts the signature but also removes the noise and other components like stamps present in the signature region. It removes the small noise components from the cheque image. After removal of all the components from the image, the signature is cropped from the cheque image to get the clean, component-removed signature image.
The techniques described herein may be used in many different operating environments and systems. Multiple and varied implementations are described below. An exemplary implementation of the system and method of the present

invention is discussed in the following section with respect to the accompanying figures.
EXEMPLARY METHOD
Exemplary methods for extraction of user-handwritten data from a document are described with reference to figures 1 -9. The methods are illustrated as a collection of blocks in a logical flow graph, which represents a sequence of operations that can be implemented in hardware, software, or a com height analysis, aspect ratio, density and histogram analysis action thereof. The order in which the process is described is not intended to be construed as a limitation, and any number of the described blocks can be combined in any order to implement the process, or an alternate process.
Additionally, individual blocks may be deleted from the process without departing from the spirit and scope of the subject matter described herein.
The exemplary method deals in the extraction of the signature from the cheque image without any knowledge of the bank's template or the user's signature. It not only extracts the signature but also removes the noise and other components like stamps present in the signature region. It removes the small noise components from the cheque image. After removal of all the components from image, the signature is cropped from the cheque image to get clean, component - removed signature.
The bank cheques bearing the user's signature is scanned. The system defines a Most Probable Signature Region (MPSR). This is the region on a cheque, which has the maximum probability of containing the signatures. Correct identification of MPSR is critical to successful extraction of the signature. Cheque sizes in different countries and banks may vary, and therefore the MPSR region may lie in relatively different locations. Therefore MPSR should be large enough to capture the complete signature but small enough to minimize capturing other

regions. Based on experimentation results, we define this region as lying between the bottom line of CAR region, the upper portion of MICR region and at X left distance from the rightmost part of the cheque. Here X may vary with cheques of different banks. Usually it varies between 2 - 2.5 inches. Once the MPSR is detected, the region is analyzed for removal of noise and other unwanted data.
Figure 1 is a flow diagram illustrating the proposed method for automatic extraction of signature image.
Referring to figure 1 automatic extraction of the signature image [101] from the document starts with the scanning of the document [102]. At block [103] a Most Probable Signature Region [MPSR] is identified and defined within the document. Noise is removed from MPSR at [104]. The resultant cheque image is binarized at block [105]. Thereafter, stamp is detected from noise removed image [106]. At block [107] stamp is extracted from noise removed image. Unwanted components attached to the signature are removed [108]. The required signature image is obtained at block [109]
Figure 2 illustrates a flow diagram representation of an exemplary implementation of the removal of noise from the signature image.
At block [201] the MPSR image is taken as input for noise removal. Thereafter at block [202] small noise components are removed from the MPSR image. This is done using low pass filtering using 3-D convolution at block [203]. At block [204] the resultant image is binarized i.e. digitized while preserving its signature components. The noise removed image is obtained at block [205].
Figure 3 illustrates the flow diagram representation of the scan line component approach.

Referring to figure 2, in scan line component approach [301], firstly a horizontal line is drawn from the bottom most area of MPSR [302]. All the components lying along this horizontal line are connected and counted as one [303]. At block [304] the line so drawn is removed and all components are restored to their original state. At block [305] a 2nd horizontal line is drawn which is a pixel above the bottom most line. All components along this line are connected [306]. Thereafter total no. of components in the image is counted at [307]. At [308] point where there is a sudden significant increase in the no. of components is marked as top of stamp. At [309] point where there is a sudden significant decrease in the no. of components is marked as bottom of stamp. Region between bottom point and top point is marked as stamp region [310]. Remaining stamp components are removed from MPSR at [311]. Image is cropped from the document at [312].
Figure 4 illustrates in detail a flow diagram representation of exemplary implementation of the removal of the components from the signature image.
Component removal [401] is usually started by carrying out Height Analysis [402] of the signature image. In this technique of component removal an analysis of the height of the signature is done and information is gathered for the past k pixels where max (y (p-k: p)), here p is position, k is width, and y is the height of columns. The procedure is repeated for the vertically flipped image as well to obtain the upper and the lower coordinates' features.
Thereafter aspect ratio [403] is calculated for all the components in MPSR region. This method is dependent on the fact that the aspect ratio of different components is noticeably different. E.g. Aspect ratio of printed text is quite high while it is relatively low for handwritten text.

A check for aspect ratio [403] is followed by Density analysis [404]. Density of the signature region, i.e. count of pixels/area of the rectangle, is calculated. If this ratio or the density value is greater than a predetermined threshold value, the component can be easily removed.
Thereafter a histogram analysis [405] of the signature image is performed. Histogram analysis of the image gives a clear idea and easily differentiates between the machine-printed text and the handwritten text. The histogram of various components is taken and analyzed for the peaks and high-density regions for component removal.
Still referring to figure 4 histogram analyses is followed by a check for overlapping region [406]. It is to be noted that existing methods are not able to extract the signatures and other user-entered information from the cheques due to the problem posed by varied backgrounds and the overlapping/intermixing of machine printed and handwritten information.
For this reason following criteria is adopted. The ratio of the overlapping region to the height of the component should be greater than a threshold for the successful removal of the component.
Thereafter results are analyzed and components are removed [407].
After the successful removal of noise and other unwanted components such as noise, a clear signature image is obtained [408].
Figure 5 a) shows a sample signature image before the removal of noise and unwanted data. Similarly, figure 5 b) shows a sample signature image after the removal of noise and unwanted data.

Figure 6 a) shows a sample signature image stamp before removal of noise but stamp region touching the signature. Similarly Figure 6 b) shows the clean signature image after extraction of stamp from the noise removed image.
Figure 7 b) illustrates a sample signature image after its detection and extraction from the MPSR region of cheque. Figure 7 a) illustrates a sample signature image with associated stamp and noise as unwanted data.
Figure 8 a) is showing a portion of a sample cheque where the signature overlaps the MICR region. Figure 8 b) illustrates an exemplary signature image with noise removed. Figure 8 c) illustrates an exemplary signature image with noise and stamp both removed.
Figure 9 illustrates an exemplary system for automatic extraction of noise free image from a document. Referring to figure 9, Input means [901] receives user inputs to customize data extraction from the given document. Acquisition means [902] reads in data from the document. Detection unit [903] detects the presence and location of the required field on given document .Memory unit [904] stores data as it is being processed. Extraction unit [905] extracts required information from the given document and the Output means [906] coupled provides the extracted information.

We claim:
1. A template independent method for automatic extraction of noise free
signature image from a document, said method comprising the steps of:
a. scanning the document to obtain a cheque image;
b. identifying and defining a Most Probable Signature Region [MPSR]
within the document;
c. removing noise from the MPSR;
d. detecting and extracting stamp(s) from noise removed image;
e. removing other unwanted components in the MPSR ;
f. obtaining the required signature image.
2. The method as claimed in claim 1 wherein the detection and extraction of stamp from noise removed image is done using scan line component approach.
3. The method as claimed in claim 2 wherein the scan line component approach comprises the steps of:
a. starting from the bottom-most area of MPSR and drawing a
horizontal straight line across the MPSR;
b. connecting all components lying along the line;
c. counting total number of components in the image;
d. removing the line drawn across the MPSR and restoring all
components to their original state
e. drawing a horizontal line a pixel above the last drawn line and
executing above sequence of steps until the entire MPSR region is
covered line by line along Y axis;

f. marking the point where sudden, significant drop in the number of
components is encountered as the bottom of the stamp;
g. marking the point where sudden, significant increase in the number
of components is encountered as the top of the stamp;
h. marking region between the bottom point and top point as stamp
region; i. removing remaining stamp components from the MPSR using
techniques including overlapping region, height analysis , density
analysis and histogram analysis and; j. cropping image from the document to get a clean signature image
for further processing
4. The method as claimed in claim 1 wherein the MPSR region lies between the bottom line of CAR [Cheque Amount Region] region, the upper portion of MICR region and at a certain left distance from the rightmost part of the document.
5. The method as claimed in claim 1 wherein unwanted components attached to the signature are removed using at least one technique including height analysis, aspect ratio, density and histogram analysis.
6. The method as claimed in claim 1 wherein the step of filtering comprises low-pass filtering with 2-D convolution.
7. The method as claimed in claim 1 wherein binarization is performed to preserve the signature image.
8. A system for automatic extraction of noise free image from a document comprising:
a. input means for receiving user inputs to customize data extraction from the given document

b. acquisition means coupled with the input means for reading in data
from the document
c. detection unit for detecting the presence and location of the
required field on given document
d. memory unit coupled with detection unit for storing data as it is
being processed
e. extraction unit for extracting required information from the given
document and an
f. output means coupled with extraction unit for providing the
extracted information
9. The system as claimed in claim 8, wherein the acquisition means includes one or more digital input devices and one or more imaging devices.
10.A computer program product for extraction of noise free signature image from a document, comprising one or more computer readable media configured to perform the method as claimed in any of the claims 1-9.