Hybrid Image Watermarking Using Local Complexity Variations Shekoofeh Azizi, Majid Mohrekesh, and Shadrokh Samavi Department of Electrical and Computer Engineering, Isfahan University of Technology, Iran Abstract— Watermarking is a method for protection of digital media intellectual property. Production of multimedia content has increased in recent years and protection of copyright is highly demanded. Different watermarking methods have been proposed to satisfy the tradeoff between imperceptibility while maintaining a proper capacity for the embedded data. In general, transform domain watermarks, such as Contourlet-based methods, produce superior results as compared to spatial algorithms. Embedding in visually more complex regions of image would enhance imperceptibility of the method. In this paper we introduce a new complexity measure for this purpose. To show the effective of this criterion, we proposed a hybrid embedding scheme based on Contourlet Transform (CT) and Discrete Cosine Transform (DCT). This is done by analyzing complexities of image blocks in the CT domain to adaptively change the watermarking strength factor. This property enables our method to take advantage of all image blocks to elevate the embedding capacity while preserving imperceptibility. Experimental results show the efficiency of the proposed method to be higher than similar works in this domain. Keywords—blind watermarking; contourlet transform; DCT; copyright protection; robustness. I. INTRODUCTION With the ease of access to internet in recent years, sharing digital media has become easier and faster. In the network environment users are capable of sharing copyrighted contents, in occasions, unintentionally. Many watermarking techniques have been proposed as a solution for copyright protection. Digital image watermarking techniques embed the owners’ copyright information into the image in embedding and extraction phases [1]. The efficiency of these techniques relies on perceptual invisibility and watermark robustness against intentional and unintentional attacks. It is essential to note that concurrently satisfying both robustness and invisibility requirements is a challenging problem. Watermarking techniques can be classified into different categories. They may be grouped into blind and non-blind techniques based on using of the original image during the watermark extraction phase. Also, they may be divided into spatial and transform domain techniques. Spatial domain methods hide information by disturbing the spatial characteristics and frequency domain methods change the coefficients of image’s transform and watermarked image obtained by taking the inverse transform. The simplest method in the spatial domain is LSB that embeds the watermark by modifying the least significant bits of pixels in an image [1]. Spatial domain methods have the advantage of low complexity and ease of implementation but are generally fragile when attacked. Since transform domain watermarking methods are more robust than spatial domain methods, in recent years several transform domain image watermarking algorithms have been proposed. Most of these techniques usually employ Discrete Cosine Transform (DCT) [3][2], Discrete Fourier Transform (DFT) [3], Discrete Wavelet Transform (DWT) [4] and Contourlet transform (CT) [5][6][7]. Among the transform domain techniques, DCT and DWT based techniques are more popular. DCT based methods are robust against simple image processing attacks and JPEG compression but unfortunately these methods are not robust to basic transformations such as cropping and resizing [1]. Since DWT has a number of advantages over DFT and DCT, it is widely used in watermarking algorithms. In general, DWT based methods use middle or high frequency regions for embedding of the watermark [4]. Despite these advantages, DWT has some limitations in capturing the directional information which is addressed by contourlet transform. In the recent years, impressive properties of contourlet transform have motivated researchers to apply this transform for watermarking purposes. Contourlet transform can efficiently capture an image’s smooth contours and edge information in all directions [8]. In [6], a non-blind CT based method for image watermarking is proposed. This method embeds the watermark into pixels selected high frequency coefficients and the number of these coefficients is related to the size of the watermark. Moreover, authors of [5] note that CT based methods outperform DWT and DCT based techniques. Song et al. [6] propose a contourlet-based image adaptive watermarking scheme in which the watermark is embedded into the largest detail subband of the image. The method presented in [7] is a non-blind method that embeds a watermark in two scales of contourlet transform. In some recent studies a combination of frequency domain transforms are used to increase robustness of watermarking schemes [9][10][11][12][13]. In [9] a DCT- DWT domain method has been proposed. This method is a dual watermarking scheme based on the orthogonal components of image sub-spaces which provide a robust