LOG-grid based off-line signature verification system

This paper presents a LOG-processed local feature based off-line signature verification system. The proposed approach has three major phases: Preprocessing, Feature extraction and Classification. The contour of the signature is divided into equal grids of size M M and the 4-directional chain code histogram of each grid is extracted, thus fed into laplacian of Gaussian filter to enhance the feature representing the signature image. These enhanced feature vectors are fed as an input to the multi-layer-perceptrons (MLP) for training purpose. The MLP is used as a recognition tool and trained with different number of training samples including genuine, skilled and random forgeries and hence tested. This model was successfully tested on two datasets, namely, CEDAR, a publicly available English signature dataset and MUKOS, a regional language (Kannada) dataset and compared with well-known approaches to exhibit the performance of the proposed approach