Abstract: The encoding step in fractal image compression is a computationally intensive operation. A large number of sequential searches through a list of domain blocks are carried out while trying to find the best match for a range block. The design of efficient search strategies has consequently been one of the most active areas of research in fractal coding, resulting in a wide variety of solutions. In this paper a review of these techniques is presented, and then a new algorithm is introduced to reduce the search complexity. When searching for the best domain block for a given range block, an initial guess at the best domain block provides the starting point. Then by invoking triangle inequality, the distance between the current best domain block and the range block constraints the search to domain blocks within an annular area about the origin. The proposed algorithm is implemented and applied to a number of test images and shows significant speedup over similar methods.
Abstract: Fractal image compression is a class of image compression techniques that offer the advantages of high compression ratio and fast decoding. But the major obstacle of these techniques is the relatively long time needed for encoding the image due to the requirement of search for the best fitting domain block for each range block. Various methods have been suggested to overcome this problem without much loss of image quality. In this paper, such methods are reviewed and a coding scheme is suggested in which multiple domain blocks are used to approximate each range block. These domain blocks are around the range block and hence a reduction in search-overhead is achieved in the encoding process. A comparative study between our approach and other fractal coding techniques is also presented.
Abstract. In this paper, a robust watermarking scheme is proposed to embed a watermark in the detailed sub-bands of a directional filter bank (DFB) decomposition of an image using a quantization process. The proposed approach uses the DFB to overcome the lack of directionality associated with discrete wavelet transform (DWT). Thus, it achieves more robustness than DWT-based methods. The algorithm starts by generating a binary logo watermark which is permuted and embedded by using the blind self image logo watermarking (SILW) algorithm. The robustness of the proposed algorithm is verified against a variety of attacks including watermark removal and synchronization removal attacks. The proposed scheme is compared with a DWT-based blind SILW. The results show that the directional frequency domain gives better robustness under similar embedding conditions than wavelet domain.
Abstract. Sub-surface and buried landmines, with the surrounding environment constitute a complex system with variable characteristics. Infrared thermography techniques are attractive candidates for this kind of applications. They can be used from a considerable standoff distance to provide information on several mine properties, and they can also rapidly survey large areas. This paper presents a robust method for landmine detection and recognition. It uses the mean-shift algorithm to segment the acquired infrared image. The segmented image retains pixels associated with mines together with background clutters. To determine which pixels represent the mines, a second phase of segmentation is applied to the output of the mean-shift algorithm by using a self-organizing maps (SOM) algorithm. Depending on the resulted cluster intensity variations, the chips extracted from the segmented image are processed to extract mine signatures. After that, the extracted signatures are scanned horizontally, vertically, and diagonally to build a cluster intensity variation profile. This profile is statistically compared with the known mine signature profiles v.The proposed system is applied on series of time variant mid-wave infrared images (MWIR), and the test result show that the system can effectively recognize the mines with low false alarm rate.
Abstract. Interest point detection and matching is a basic computer vision task. This paper uses the Non-Subsampled Contourlet Transform (NSCT) detector with some local descriptors to develop a robust Interest point matching system. The NSCT-based detector is very efficient in extracting relevant image features that have good localization and rich geometric information. Once interest points have been extracted, invariant descriptors can be computed to represent these points and match them by comparing their descriptors using similarity distance metric measure. Several experiments have been conducted on a variety of datasets that compare the proposed approach to the widely used Harris and Hessian detectors with the same local descriptors. The results show that the NSCT method is efficient in natural scenes with distinctive texture, occlusion and clutter objects. It offers robustness for detecting points in blurred images. It is also invariant to rotation, translation, and partially invariant to scale and viewpoint angle changes.