Multi-scale Local Binary Patterns- A Novel Feature Extraction Technique for Offline Signature Verification

Abstract

This paper presents a powerful feature representation method called Multi-scale Local Binary Patterns for offline signature verification. The multi-scale representation oriented local binary patterns can be obtained by changing the radius R value of Local Binary Patterns(LBP) operator and combining the LBP features at different scales. In this proposed approach the LBP operator is applied at 3 different scales by varying the radius R value and at each scale equal number of pixels are considered for the processing. Finally, by cascading a group of LBP operators at 3 different scales over a signature image with fixed number of pixels at each scale and combining their results, a multi-scale representation LBP can be obtained. This essentially represents nonlocal information. Features fusion is performed by the linear combination of the histogram corresponding to 3 different radii results in a multi resolution (scale) feature vector. Support Vector Machine (SVM) is a well known classifier employed to classify the signature samples. Experimental results on standard datasets like CEDAR and a regional language datasets shows the proposed technique’s performance. A comparative analysis with few well known methods is also presented to demonstrate the performance of proposed technique.