Efficient Image Watermarking Using Filtered DWT-Blocks for Quantization of Significant Differences

In the paper, a robust blind watermarking method is introduced for gray-scale images based on wavelet tree quantization with an adaptive threshold in the extraction. Every block of 2× 2 coefficients of High-Low subbands of the Wavelet tranform are grouped in a block through the parent-child relationship of the wavelet tree. Every scrambled binary watermark bit is embedded into each block based on the difference value of two largest coefficients. The watermark is recovered by comparing the difference values in each block to an adaptive threshold. The accuracy of an extracted watermark depends on the threshold which is determined by minimizing the sum of weighted within-class variance. The performance of the proposed watermarking method is represented through experimental results under various types of attack such as, Histogram Equalization, Cropping, Low-pass Filtering, Gaussian noise, Salt & Pepper noise and JPEG compression. In additions, the proposed method is also compared to recent methods in the extraction performance. Keywords– blind watermarking, 3-level DWT, wavelet tree, significant difference, adaptive threshold, within-class variance.