Image Forgery Detection Using Multi-Resolution Weber Local Descriptors Muhammad Hussain 1 , Ghulam Muhammad *2 , Sahar Q. Saleh 1 , Anwar M. Mirza 2 , and George Bebis 3 1 Department of Computer Science, 2 Department of Computer Engineering College of Computer and Information Sciences, King Saud University, Riyadh 11543, Saudi Arabia 3 Department of Computer Science and Engineering, University of Nevada at Reno, USA * ghulam@ksu.edu.sa Abstract— In this paper, a multi-resolution Weber law descriptors (WLD) based image forgery detection method is introduced. Due to the maturing of digital image processing techniques, there are many tools, which can edit an image easily without leaving obvious traces to the human eyes. So the authentication of digital images is an important issue in our life. The proposed multi-resolution WLD extracts the features from chrominance components, which can give more information that the human eyes cannot notice. A support vector machine is used for classification purpose. The experiments are conducted on a large image database designed for forgery detection. The experimental results show that the accuracy rate of the proposed method can reach up to 93.33 % with multi-resolution WLD descriptor on the chrominance space of the images. Keywords: image forgery detection, Weber local descriptor, splicing forgery, copy-move forgery, multi-resolution method I. INTRODUCTION Nowadays, we are living in an age, where digital imaging has grown and developed to become the widespread technology. It plays a significant role in human life, where the digital images are currently allowed as official documents that can be used in daily newspapers, magazines, military, and may be used as proof at court or in the medical diagnose field [1]. With the increasing applications of digital imaging, different types of software are introduced for image processing. Such software can do an alteration in digital image by changing blocks of an image with no showing the effect of the modification in the forged image. These modifications cannot be noticed by human eyes. The image forgeries can wipe off an important object from a proof image which can be a reason for miscarriage the court. Therefore, the security of digital images is a very important topic of research. Many techniques have been developed for authenticity guarantee of the digital images. These techniques can be divided into intrusive (active) and non intrusive (blind or passive). The active techniques can be classified into two categories. The first one needs to embed a watermark in the image, while the second is a digital signature-based technique. In these methods, particular data is embedded in the digital images for supporting multimedia digital authentication and rights safety. If the image contents have been modified, the embedded data will be changed too. The image authenticity is done by checking whether the true signature corresponds to the signature that is retrieved from the suspicious test image. These techniques are restricted because of the inability of many digital cameras to embed the signature. Due to the restrictions of active techniques, which involve pre-processing in order to create the labeled images, the researchers tend to develop non intrusive techniques for validating the authenticity of digital images. These techniques examine images with no embedded data such as signatures or watermarks, and result whether these images are authentic or tampered. The most commonly used forgery is copy-move forgery (CMF), where a region of an image is copied and moved to another region in the same image to make region-duplication in order to conceal an important object from the original image. The copied block may be changed by any kind of preprocessing procedures such as rotation, scaling, additive noise etc. to suit the copied area with the whole image. In another type of forgery, one part of an image is copied and pasted it in another image. This type of forgery is called splicing. In this paper, we propose a multi-resolution Weber law descriptor based method to detect image forgery. The forgery can be either copy-move or spliced. The proposed method is evaluated on different types of post-processing in forgery on a large image database. The rest of the paper is organized as follows. Section 2 reviews some related previous work, Section 3 presents the proposed image forgery detection method, Section 4 gives experimental results with discussion, and finally, Section 5 draws some conclusion. II. RELATED PREVIOUS WORK An improved DCT (discrete cosine transform)-based technique was proposed in [2] to discover CMF in digital images without any previous information of the suspicious image. The image is subdivided into blocks and the DCT is computed. The DCT coefficients are lexicographically sorted, and compared with different blocks. The proposed technique is robust against JPEG compression, additive white Gaussian noise, or blurring distortion. Cao et al [3] proposed an improved DCT-based method to locate the duplicated regions in a given image. The method uses circle block for representing DCT coefficient's array. Noise pattern based image forgery detection method was proposed in [4]. Noise pattern is obtained by subtracting the EuroCon 2013 • 1-4 July 2013 • Zagreb, Croatia 1570 978-1-4673-2232-4/13/$31.00 ©2013 IEEE