An Image Watermarking Scheme using HVS Characteristics and Spread Transform Santi P. Maity Bengal Engineering College (DU), Dept. of Electronics and Telecommunication Engg. P.O.-Botanic Garden, Howrah 711 103, India spmaity@telecom.becs.ac.in Malay K. Kundu Indian Statistical Institute Machine Intelligence Unit 203 B. T. Road, Kolkata 700 108, India malay@isical.ac.in Abstract The paper presents a robust digital image watermark- ing scheme that uses both the characteristics of the human visual system (HVS) and statistical information measure. Spread transform approach is used where data is embedded through transform coeff icients of both the cover and the watermark data. The spread transform watermarking tech- nique yields better results in terms of imperceptibility, re- siliency, capacity and cost compared to widely used spread spectrum watermarking schemes. Hadamard transforma- tion is used not only for simpler implementation but also for its higher data hiding capacity [1]. Experimental re- sults show that the visual quality of the extracted watermark is good in spite of several external attacks. The fact is also supported by mutual information values used as objective measure. 1. Introduction Recently watermarking is used widely for ownership protection, authentication, and content integrity verification of intellectual property in digital form. Several watermark- ing techniques for digital images have been proposed in the literature where the characteristics of HVS (human vi- sual system) further improves the imperceptibility and ro- bustness performance of traditional spread spectrum wa- termarking schemes. Cox et.al proposed a global DCT (Discrete Cosine transform) based spread spectrum water- marking [2] and Podilchuck and Zeng proposed perceptual model-based watermarking scheme using DCT and Wavelet transform [3]. The embedded watermark signal in both cases consists of a sequence of real numbers that are nor- mally distributed but does not convey unique signature. Hence, their detection methods depend on similarity mea- surement that needs the inevitable presence of watermark signal at the receiver. We use spread transform watermark- ing scheme in order to achieve better imperceptibility and resiliency, although our spread transform scheme is differ- ent to that of Chen and Wornell scheme [4]. In our scheme, transform coefficients of the watermark image modulate the corresponding significant transform coefficients of the cover image according to the Watson model of HVS [5]. We use meaningful gray scale image as watermark so that it not only conveys a unique information but also shows a good degree of resiliency after various forms of image distortion. Hadamard transformation used offers low computation cost and higher data hiding capacity at low quality factor com- pression [6]. Although our decoding method is based on objective measure like mutual information, but subjective quality of the extracted image is also visually acceptable. 2. Watermark embedding and decoding We use in this work gray scale image for both the cover and the watermark image. 2.1. Watermark embedding Step I: Spatial dispersion of watermark image The watermark image is spatially dispersed using a cryp- tic key (k) generated by linear feedback shift register. The process converts gray scale watermark image into noise-like image and thus increases imperceptibility by spreading wa- termark information over wide region of the cover image. Step II: Image transformation The block (8 ×8) based Hadamard transformation is applied over both the cover and the spatially dispersed watermark image. Step III: Image dependent permutation In order to increase imperceptibility, the transform co- efficients of the cover and the watermark image are sorted in ascending order. Step IV: Generation of modulation function We construct modulation function based on Watson visual [5] and entropy masking model. Watson relates frequency sensitivity (F u,v,b ), luminance masking (L u,v,b ) and con- trast masking (C u,v,b ) with the DCT coefficients according 0-7695-2128-2/04 $20.00 (C) 2004 IEEE