Font Size: a A A

Research On Natural Images' Processing Theory And Method

Posted on:2005-02-01Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y S TanFull Text:PDF
GTID:1118360152457203Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
Image has been widely used to communicate with each other. There are various applications and complex characteristics of images. Therefore, computer image processing technology is a very hot focus. Many researchers have devoted themselves to it and gotten some exciting results. Meanwhile, there are still a lot of problems remained.Comparing with artificial images, natural images are more valuable. But it is difficult to process natural images by traditional processing methods because natural images have some complex characteristics, such as non-Gaussian, self-similarity, scale invariance and high dimensional singularity. The dissertation deals with these problems and tries to propose some systematic and efficient mathematic tools and engineering solutions for natural images processing.Analyzing the native characteristic of natural images and the essential of traditional processing technologies, the dissertation draws a conclusion that the primary drawback of traditional mechanisms is that traditional ones can only describe the low dimensional singularity (point). However, natural images have a lot of high dimensional singularity signal. Therefore, the dissertation introduces the concept of the Generalized Singularity and uses it to describe the high dimensional singularity of natural images.The dissertation introduces the simple Radon transform. Moreover, it gives the definition of the Generalized Radon Transform, its properties and the discrete form. Since one practicable transform should have its inverse transform, the dissertation proves the inverse Generalized Radon Transform with the common transform reference function.Within natural images, the high dimensional singularity signal is the edge. Therefore, efficient edge processing methods are the foundation of post-processing methods. The dissertation argues that one systematic edge processing flow should include three parts, i.e. edge detecting, thinning and linking. The dissertation has introduced the traditional edge detecting and linking methods. Meanwhile, it presents a novel edge thinning algorithm named as Connectivity Preserved Edge Thinning Algorithm, which combines the detecting strategy of the connectivity critical point with the estimating method of the edge direction. The CPETA algorithm can give ideal single-pixel-wide thinning results and guarantee the image's original edge connectivity.Traditional edge linking methods cannot get good results for natural images that have many complex edges. With the help of parameter space's constraint by GRT reference function, thedissertation proposes a novel discrete Generalized Radon Transform computing strategy named as Back Transform Method. BTM can greatly reduce the GRT's computing cost. Based on BTM, this dissertation presents the GRT-based Edge Linking Integrated Method, which can detect, link and smooth natural images' edges efficiently.The computing cost of the normal GRT computing strategies is so large that the GRT cannot be applied practicably. To reduce the computing cost, the dissertation achieves two practicable improvements. First, since the GRT is linear and addible, the dissertation proposes the parallelized implementation of Generalized Radon Transform. The implementation is based on the GRT Cartesian space's decomposition. It makes the GRT more practicable. The second method modifies the transform object from point to point set. A novel Generalized Radon Transform method is proposed based on this new idea. With the new method, the dissertation presents the GRT's decomposition strategy and analyzes its complexity.Combined with natural image processing technologies presented, the dissertation proposes an efficient natural images processing framework named as Integrated Image Processing Abstract Framework. The IIPAF uses the Generalized Radon Transform as the pre-processing method, and other post-processing methods deal only with the middle image data that is GRTed. Experiments prove that the IIPAF can represent and process natural images efficiently.Since natural images are non-Gaussian, self-s...
Keywords/Search Tags:Natural Image, Generalized Singularity, Generalized Radon Transform, Edge Thinning, Edge Linking, Image Fractal Compression, XLet
PDF Full Text Request
Related items