Face Template Security: LBP-Based LSB Watermarking Technique for Multi-class SVM Classification Using HoG

Abstract

A digital watermarking is a technique which is protecting the user authenticity by embedding the digital code in the image. In this paper, an attempt is made to invisible watermarking technique based on Local Binary Pattern (LBP), Least Significant Bit (LSB), Gabor, Histogram of Gradient (HoG), and Multi-class Support Vector Machine (SVM). The embedded image is classified through multi-class SVM with magnitude and direction by HoG features. It is a kind of invisible watermarking technique as we need the original image for identification. The Peak Signal Noise Ratio (PSNR) 31.12 dB is used to evaluate the strength of noise between the watermarked image and original image. Subsequently, the image identification is done by using multi SVM classification and measured with 98.75% accuracy of watermarked image.