A new robust image watermarking approach using two-level intermediate significant bits coupled with histogram intersection technique

Abstract

Nowadays, due to the explosive growth of the internet and multimedia technologies, digital multimedia assets are mostly vulnerable to redistribution and replication via accessible networks. Hence, digital watermarking techniques have become widely recognized as effective solutions for ownership identification and copyright protection of the digital assets. Experimental investigations have shown that current digital image watermarking approaches are prone to be impacted by watermarking attacks which originated from image processing or signal processing common operations. Even worst, some attacks with unknown or complex behaviours can be emerging in near future that able to destroy partially or completely the embedded watermarks. In such a situation, providing a universal model for the watermarking attacks is almost impossible. Beside the robustness, it is important to provide high visual quality and embedding capacity. Therefore, creating an effective watermarking technique which provides a balanced trade-off between robustness and visual quality, and at the same time attain reasonable capacity is a challenging task. To take up this challenge, a two-level Intermediate Significant Bit watermarking technique called BiISB is introduced in which two interrelated watermarks namely, main watermark and sub-watermark which is formed from statistical information of the main watermark using binary bit-patterns, are embedded concurrently in the host image. After an attack, the remnant information of both main watermark and sub-watermark in the form of bit-pattern histograms are used for the ownership identification of the property using Histogram Intersection technique. In addition, in order to measure the trade-off among above requirements, two techniques namely, Threshold based approach and Fuzzy approach are introduced. Experiments have been conducted using arbitrary watermarks, 10 standard host images, 10 different attacks, and 10 different embedding capacities. The experimental results revealed that the proposed technique successfully identified the ownership of the watermarked images even after the embedded watermarks were totally corrupted. The results also revealed that the technique introduced has successfully balanced the trade-off between robustness and quality, and at the same time attained high capacity. This is realized by obtaining ownership probabilities of higher than 0.95, qualities beyond 40dB, and 12.5% capacities.