With the development of computer hardware and software, and the gradually maturity of related disciplines, such as pattern recognition, image processing and computer graphics, the development of computer vision is also in full swing, and becomes the current research focus. In recent year, a large number of computer vision problems have been proposed and deeply studied by researches. For many vision problems, due to their complexities, it is difficulty to utilize a global model to represent them. Hence, it is more flexibly and effectively to address vision problems by considering them locally, that is, introducing local perception. In this thesis, we apply the localization idea to some vision problems, and correspondingly propose some local perception based algorithms. The main contributions of the thesis are listed as follows: 1. We propose a local background-aware visual tracking algorithm. First, the proposed tracking algorithm treats local background as the context, and introduces it into target representation. As a result, we propose a target description, i.e. the local background weighted histogram (LBWH). The LBWH enhances the discrimination between target and background, so that highlights the foreground in the target region. Hence, the proposed tracking algorithm can effectively tackle the difficulties of the similarity between target and background as well as the variability of the potential background. Second, we propose a forward-backward mean-shift (FBMS) algorithm is proposed by incorporating a forward-backward evaluation scheme, in which the tracking result is determined by the forward-backward error. Therefore, the proposed algorithm can effectively solve short-time partial occlusion and illumination variation problems. Extensive experiments on various scenarios have demonstrated that our tracking algorithm outperforms the state-of-the-art approaches in tracking accuracy. 2．We propose an edge-directed single image super-resolution method based on a local gradient magnitude adaptive self-interpolation (LGMASI) algorithm. Essentially, the LGMASI algorithm utilizes the local gradient information to adaptively sharpen the gradient. An adaptive self-interpolation algorithm is first applied to estimate a sharp high-resolution gradient field directly from the input low-resolution image. The obtained high-resolution gradient is then regarded as a gradient constraint or an edge-preserving constraint to reconstruct the high-resolution image. Exte...
修改评论