Robust digital image watermarking using LWT and Random-Subspace-1DLDA with PCA based statistical feature reduction

Abstract

The proposed work presents a novel ensemble model based blind image watermarking using a random-subspace-one dimensional linear discriminate machine learning model for providing digital security to images. Security of digital content is one of the important concerns in today’s era. In this work, the extraction of a binary watermark is considered as the binary (0 or 1) classification problem. The good learning rate of the linear discriminate analysis model and good performance of ensemble-based machine learning models motivates this investigation because it has a good ability to learn the relationship between the position of the watermark bit and its neighbor in the given image. Here two types of watermark reference and a signature watermark are used for the training and the testing purpose. The reference watermark bit is generated randomly and the signature watermark bit is original watermark. The watermarking scheme is tested against most of the image attacks and compared with some of the similar existing methods that have used SVM for watermark extraction and found robust against most of the image attack such as Median filter, Salt and pepper noise, Speckle noise, JPEG with quality factor 50% and Crop(25%) attack. It also achieves an imperceptibility of 35. 47dB for Lena image, 35. 14dB for Mandril image, and 35.51dB for Peppers standard image in case of no attack.