Robust and low cost watermarking using image characteristics Santi P. Maity Malay K. Kundu Prasanta K. Nandi Dept. of E&TC Engg. Machine Intelligence Unit Dept. of CS&Tech. B. E. College (DU), Howrah Indian Statistical Institute, Kolkata B. E. College (DU), Howrah spmaity@telecom.becs.ac.in malay@isical.ac.in pkn@cs.becs.ac.in Abstract Most of the digital image watermarking techniques use pixel values, frequency or other transform coefficients to embed information without considering the perceptu- ally significant portion of the cover. The present work selects the perceptually significant region of the cover and embeds data in the transform coefficients in order to design low-cost robust watermarking scheme. Exper- imental results using several benchmark image samples are reported. 1. Introduction Advancement in digital techniques and rapid expan- sion of the Internet have created the need of ownership protection, authentication and content integrity verifi- cation of intellectual property etc. and the objectives are fulfilled using digital watermarking [6]. The essen- tial requirements of digital image watermarking are im- perceptibility, robustness, security of the hidden data, embedding rate i.e. capacity, complexity and compu- tation cost etc. Several watermarking schemes for digi- tal images have been proposed in the literatures where data are embedded directly in pixel values or in fre- quency or in other transform coefficients of the cover, in order to meet these requirements [5] . Robustness requirement of image watermarking is achieved if data is embedded in the region bearing es- sential characteristic information of the image such as edges, texture and high gray level curvature points etc [7]. This is due to the fact, so long different characteristics regions of an image are not drastically changed, hidden data can be extracted faithfully. The present work selects the watermark embedding region based on an edge entropy measure of image block. Data embedding in the low edge blocks of the cover image is resilient against lossy compression but leads to a large degree of image visual distortion. On the other hand, distortion due to data embedding in the high edge blocks, is less visually perceivable but hid- den information might be lost after lossy compression attack. Hence, an imperceptible and compression re- silient image watermarking can be achieved if medium edge blocks of the cover image is selected. Resiliency is increased further if suitable transform coefficients, rather than pixel values of the regions, are used for data embedding [8]. Transform domain approach in- creases the computation cost of data embedding and recovery over spatial domain schemes. The selection of proper transformation, e.g. Walsh, Hadamard etc. reduces the degree of computation cost. The present work embeds data in the suitable Walsh coefficients of the medium edge blocks. The watermark embed- ding regions are selected using edge entropy value of a block, as discussed in section 3.1, so as to achieve a good compromise between robustness performance and quality of the embedding process. 2. Image transform The forward (Equation 1) and inverse (Equation 2) kernels of discrete Walsh transform are identical with signed integer value and are given as follows g(x, y, u, v)=1/N n-1  i=0 (−1) bi(x)bi(u)+bi(y)bi(v) (1) and h(x, y, u, v)=1/N n-1  i=0 (−1) bi(x)bi(u)+bi(y)bi(v) (2) where b k (z) is the k-th bit in binary representation of z [3]. The signed integer valued kernel does not require floating point multiplication when convolved with digi- tal image, thus yields low-cost watermarking. The ker- nels being identical, a single hardware block can be used to implement the forward and inverse transfor- mation.