14 th International CSI conference (CSICC2009), July 1-2, 2009, Tehran, Iran A Watermarking Method Based on Optimizing SSIM Index by using PSO in DCT Domain Mohsen Rohani * and Alireza Nasiri Avanaki ** * ECE Department of Tehran University, Tehran, Iran. Email: m.rohani@ece.ut.ac.ir ** ECE Department of Tehran University, Tehran, Iran. Abstract— A watermarking method in DCT domain is modified to achieve better imperceptibility. Particle Swarm Optimization (PSO) is used to find the best DCT coefficients for embedding the watermark sequence and the Structural Similarity Index is used as the fitness function in order to have a watermarked image with the best possible quality. Keywords—Image Watermarking; Structural Similarity Index; Particle Swarm Optimization; Human Visual System I. INTRODUCTION With the significant growth of the internet and digital multimedia technology, the need of using digital watermarks in order to distinguish the original contest from the duplications, especially in the internet-based multimedia applications, is essential. Digital watermarking which involves hiding or embedding information codes such as copyright messages, ownership identifiers, bi-level or grey level images directly in signals such as audio, video, images, text and graphics is used for copyright protection, content authentication, media forensics, data blinding, broadcast monitoring, and covert communication [1]. In this paper we propose a method for finding the best coefficients for embedding the watermark to decrease its visibility in terms of SSIM. One of the important properties of the watermark is imperceptibility. The watermark should be imperceptible so the human could not be able to percept any changes in the media. There are different algorithms for image quality assessment. One of the most recent algorithms is structural similarity index which is one of the best estimations of the HVS perception. There are several methods which are applied to gain the proper imperceptibility. In these methods the main idea is to adaptively adjust the embedding strength based on HVS during the watermark embedding process. Two JND based watermarking algorithms in the DCT and DWT domain proposed separately in [2]. A watermarking algorithm by assigning different embedding strength in the high contrast areas and the directional edge areas is developed in [3]. A new spatial masking, based on the luminance masking, texture masking and the edge masking, to embed watermarking adaptively is proposed in [4]. A watermarking algorithm based on HVS is proposed in [5]. A watermarking method using Genetic algorithm for improvement of imperceptibility in DCT domain is proposed in [6]. In this paper, a new watermarking algorithm for improving imperceptibility by choosing DCT coefficients using PSO for embedding watermark is proposed. The proposed algorithm uses the SSIM index as the quality metric which takes the human visual system characteristics into consideration [7]. The paper is organized as follow. In section 2 the basic introduction to PSO and SSIM is given. In section 3 our proposed watermarking algorithm is proposed. And conclusion is given in section4. II. INTRODUCTION OF PSO AND SSIM INDEX In this section, the basic concepts of PSO and Structural Simmilarity (SSIM) image quality metric are presented. A. Basic Concepts of PSO Particle swarm optimization which is a stochastic population based evolutionary algorithm was developed by Kennedy an Eberhart in 1995 [8]. The aim of this algorithm is to simulate forging trend and communication behavior in flocks of birds when they are flying. Each bird is a particle which flies through a search space and have two characteristics: its memory of its own best position and knowledge of the global best. The particles start at a initial position and try to search for the maximum or minimum of a objective function by searching through the search space. Each particle changes its position in the search space by knowing its velocity and the position where good solutions have already been found by the particle itself or its neighboring particles. There is two kind of PSO: Global version or the local version. When the whole swarm is defined as neighborhood it is called the global version, and otherwise it is called the local version. Equations (1) and (2) describe how each particle calculates its next position and velocity. ݒ ௜ௗ ൌ ݒ .ݓ ௜ௗ ൅ ଵ ݎ. ଵ . ሺ ݐݏെ ݔ ௜ௗ ሻ൅ ଶ ݎ. ଶ . ሺ ݐݏ ௜ െ ݔ ௜ௗ ሻ ሺͳሻ ݔ ௜ௗ ൌ ݔ ௜ௗ ൅ ݒ ௜ௗ ሺʹሻ Each particle updates its velocity by using equation (1).  ଵ ,  ଶ are cognitive coefficients and ݎ ଵ , ݎ ଶ are random cumbers between [0,1];  ݐݏis the global best position and  ݐݏ ௜ is the individual best of a specific particle . And equation (2) updates each particle’s position in the search space. B. Structural Similarity Index Let ݔൌሼ ݔ ௜ | ൌ ͳ,ʹ, … , ሽ and ݕൌሼ ݕ ௜ | ൌ ͳ,ʹ, … , ሽ be two discrete non-negative signals that have been aligned with each other and extracted from the original and