Signature Verification using Edge Detection and Oc-Svm
Mokshith Reddy A.V1, Harsha Nimmagadda2, Venkatesh Pallapothu3, Jasmine Pemeena Priyadarsini.M4
1Mokshith Reddy A.V, Department of Electronics and Communication Engineering, Vellore Institute of Technology, Vellore, India.
2Harsha Nimmagadda, Department of Electronics and Communication Engineering, Vellore Institute of Technology, Vellore, India.
3Venkatesh Pallapothu, Department of Electronics and Communication Engineering, Vellore Institute of Technology, Vellore, India.
4Jasmine Pemeena Priyadarsini. M, Department of Electronics and Communication Engineering, Vellore Institute of Technology, Vellore, India.
Manuscript received on May 02, 2020. | Revised Manuscript received on May 21, 2020. | Manuscript published on May 30, 2020. | PP: 2763-2767 | Volume-9 Issue-1, May 2020. | Retrieval Number: A3121059120/2020©BEIESP | DOI: 10.35940/ijrte.A3121.059120
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: There are many researches going on in the field of Image processing to find an accurate model for detection of forged signatures. It is compulsory to select appropriate algorithm for achieving best results. Many have implemented the available algorithms like LBP, HOG, geometric features and some have designed their own extraction techniques like Histogram of template and mixture of many complex algorithms to detect the forgeries. In our model we extracted edge features of the image using canny edge detection and then extracted features using HOG and then we calculated area, standard deviation, centroid, kurtosis etc. for the edge image which give a feature vector of length 262. There is little research towards one class SVM or outlier detection. We trained our model with different kernels of SVM to find which kernel gives best result. This OC-SVM will be very helpful than a multi class SVM as we will be having only original signatures of the users and not forged so it will be best to use though it is very sensitive it can be used even in the real world.
Keywords: Average Error Rate (AER), False Acceptance Rate (FAR), Feature extraction, False Rejection Rate (FRR), Histogram of Oriented Gradients (HOG), Preprocessing, Support Vector Machine (SVM), Radial Basis Function (RBF).
Scope of the Article: Machine Learning