Weighted Extreme Learning Machine for Digital Watermarking in DWT Domain

Abstract

In this paper, a digital watermarking scheme has been implemented using Weighted Extreme Learning Machine (WELM) on images and results are compared with other existing methods. The neighbourhood relationship among the pixel in image can be used as an reference positions, WELM is used as a regressor. Digital watermarking problem can be treated as regression problem can be trained at the embedding procedure and watermark or logo or sequence can be embedded. The watermark as an information can be embedded into blue channel of input images used for watermarking taking into account of human vision system (HSV). As WELM algorithm is very fast, cost sensitive and has good learning and generalization ability, the watermark can be correctly extracted despite of the watermarked image subject to several malicious attacks. Experimental results show that the WELM based watermarking scheme outperformed other existing methods against different attacks including salt & peppers (0:04), scaling 50%, cropping 15%, rotation 150 etc. As implemented digital watermarking scheme is robust and imperceptible determined based on calculated metrics PSNR, BER.