| The main study objective of computer vision is that how to realize human being's vision function using computers, namely to recognize and understand the three-dimensional (3-D) world from two-dimensional projective images. Most common in human experience is ordinary 3-D world in which we live. All the objects in natural scene are three-dimensional in spatiality, moving in status, where stillness is relative. So the research and application on computer vision should mainly be based on the condition of 3-D state and motion. In deed, most of the 2-D images we see have been derived from this 3-D world by camera systems that employ a perspective projection to reduce the dimensionality from three to two. And, the spatial information of the 3-D object is perhaps contained hi the 2-D image, hi order to understand the nature from images, the process of getting 3-D information from 2-D images is necessary.It is very significant to study computer vision, because it is not only beneficial to satisfy the application requirements of artificial intelligence, but also to be helpful to deeply understand and study the mechanism of human being's vision system.The main study contents of Computer Vision include input device, low level vision, middle level vision, high level vision, and system architecture. In this paper, to the question of input device, edge detection as part of low level vision, stereo vision and 2-D motion estimation as parts of middle level vision, some approaches and arithmetic are studied deeply, what are also the main contents of this paper.About image capturing, based on other researchers' study, an optical-electronic imaging equipment, Ring Aperture Microscope (RAM), is designed, including its fabrication method and corresponding image restoration technique. Using the coded-aperture imaging technique, RAM is suitable for imaging in X-ray, with high spatial resolution and fast temporal resolution.About feature extraction, a new method of image edge detection, using multiple edge features and fuzzy reasoning, is submitted. To effectively extract edges, edge pixels are classified according to the difference of intensity distribution, and multiple effective edge features are defined. Edges are detected by applying fuzzy reasoning (min-max centroid method), based on local image edge characteristics. The given algorithm is robust for noise, and specially has good detection result for the unclear edge.Stereo vision studies on the technique by which one can deduce the 3-D information of an object (such as shape, range, and so on) from a stereoscopic image pair. In a complete stereo vision system, there are six modules, which are images capturing, camera adjustment, feature extraction, stereo matching, 3-D information estimation, and post-processing. Especially, the stereo matching is the difficult and important part. Here, two kind of stereo matching approaches are studied, based on improved genetic algorithms and wavelet transform respectively. Both methods have good matching results in matching experiment.Moreover, the research on Dynamic Scene Analysis is also the main study field of computer vision, in order to satisfy people's demand for learning more about the 3-D object and world. 2-D motion estimation, as the base of Dynamic Scene Analysis, is very important and worth being studied more deeply.2-D motion estimation is to estimate object motion, extract motion parameter, and analyze motion rule, based on 2-D image sequence. In order to extract the contour of active image and estimate 2-D motion effectively, a method using Snakes model, where initial contour of every frame is predicted by optical flow, is given in this paper. With this improved method, Snakes model has better performance for contour extraction of active image. Compared with standard Snake method, the given algorithm can get better estimation result.In addition, some studies on genetic algorithm are done, and a new general hybrid genetic algorithm (GHGA) that combines the local direct search algorithm with canonical gene...
|
PDF Full Text Request