Abstract—The comparison between the hybrid watermarking using genetic algorithm and the pure hybrid DCT-SVD (Discrete Cosine Transform-Singular Value Decomposition) is presented. The watermark scheme is based on DCT-SVD]; applying DCT to the host image, map the DCT coefficient in a zigzag order, then propose the SVD on the sub-block and the singular value in each sub block that will be modified to embed the watermark image. Afterward the image quality can be improved with the GA-based evolution. Experimental results are shown the best extracted watermark quality of hybrid DCT-SVD between the hybrid using genetic algorithm and pure hybrid DCT-SVD. Index Terms—Hybrid, genetic algorithm, DCT, SVD. I. INTRODUCTION Digital watermarking is a process of embedding data into a host data such as image, video or audio. The benefit of digital watermark are illegal distribution of digital data, copyright protection, copying of discourage unauthorized etc [1]. The important requirements of digital watermarking consist of robustness, unambiguity, perceptual transparency and capacity [2]. Robustness is a capability of watermark to various modifications or manipulations as well as cropping, filtering, noising, compression etc; unambiguity is umbigious after the extraction process; perceptual transparency is perceptual invisible after embedding process; capacity is put in some data into host image without loosing transparency. Watermarking scheme categories are permanence, visibility and domain. The watermark scheme based on domain transform has several advantages, these are perceptibility and robustness [3]-[5]. In this paper, we carried out the scheme based on DCT-SVD; propose DCT to the host image, mapping the DCT coefficient in a zigzag order; and then applying the SVD on the sub-block and the singular value in each sub block that will be modified to embed the watermark image. Afterward the image quality can be improved with the GA-based evolution. Our experiment use the peak signal to Manuscript received March 23, 2012; revised May 6, 2012. This work was supported in part by the DIPA Research in Indonesian Institute of Sciences. Didi Rosiyadi and Shi-Jinn Horng are with the Department of Computer Science and Information Engineering, National Taiwan University of Science and Technology (e-mail: didi.rosiyadi@gmail.com, horngsj@yahoo.com.tw). Nana Suryana and Nurhayati Masthurah are with the Research Center for Informatics-Indonesian Institute of Sciences Jl. Cisitu, Sangkuriang, Bandung, West Java, Indonesia (e-mail: nana@suryana.or.id, masthurah@informatika.lipi.go.id, Phone. +62-22-2504711). noise ratio (PSNR) to measure the quality of the watermarked image and the correlation coefficient to verify the existence of the watermark. Experimental results are shown the best extracted watermark quality of hybrid DCT-SVD between the hybrid using genetic algorithm and pure hybrid DCT-SVD. II. THE PROPOSED METHOD The proposed method of watermark scheme is introduced. The watermark scheme based on DCT-SVD; one of the advantages is durability to a variety of attacks. According to description in [6]-[10], the watermark embedding and extraction can be seen as follows. A. Watermark Embedding  Apply the DCT and Divide the host image into four quadrants ( SB1, SB2, SB3 and SB4)  Scan the DCT coefficient in the zigzag manner  Perform SVD operation for each quadrant  Perform DCT SVD operation for watermark  Insert the Watermark for each quadrant  Modified the coefficient back to their original positions  Apply the Inverse DCT to produce the watermarked host image. B. Watermark Extraction  Apply the DCT and Divide the host image into four quadrants  scan the DCT coefficient in the zigzag manner  Apply SVD operation for each quadrant  Extract the singular values for each quadrant  Construct the DCT coefficient for 4-visual watermarks C. Optimization Process by Genetic Algorithm  We embed the watermark in 4 different embedding positions. The coordinate of embedding position are g and h (g,h) in an 8 x 8 block. A coordinate g and h can represent by 6 bits and then one binary string has 24 bits, as illustrated 100111010111001010110001. For each embedding position decide the optimal value of scaling factor key.  The Initialization is a process to create a population of chromosomes and to initiate the vectors randomly for chromosomes in each block.  Fitness function value. In this stage the children chromosomes are generated in accordance with fitness function value. Then to measure the fitness function value is defined as follow : ) ) , ( ) , ( ( 1 1 * *     n i j w j j i j i I I WE W W WI n f A Comparison between the Hybrid Using Genetic Algorithm and the Pure Hybrid Watermarking Scheme Didi Rosiyadi, Member, IACSIT, Shi-Jinn Horng, Nana Suryana, and Nurhayati Masthurah International Journal of Computer Theory and Engineering, Vol. 4, No. 3, June 2012 329