Microscopic Image Feature Matching And 3d Display Preliminary Study

Micro stereo vision (MSV) is the application of stereo vision techniques in microcosmic domain. Stereo microscope and CCD sensors make the main components of a MSV system. In order to realize the 3D reconstruction and measurement of microcosmic objects and scenes, the stereo vision method is used usually. In this paper, such problems as image pre-process, character extraction, stereo matching, depth recovering, display of reconstruction result are studied.According to the characters of microscopic images, several methods are used to decrease the noise of image and to increase the image contrast in pre-process. The objective area and the background area are separated well and the quality of image is increased obviously after pre-process. Based on the basic theory of character extraction, the edge detection operator is chosen to act the role of special points extraction operator. In order to choose an effective operator, several operators such as Sobel, Laplacian, LoG and local entropy of edge detection are used. A new algorithm based on edge sampling and entropy curve analysis is brought forward through the improvement of the local entropy algorithm. The operators of LoG and local entropy are selected to extract the special points of stereo image. The primary analysis is done to the extraction result.Stereo matching is the emphasis of this paper. Several matching elements such as gray scale, special points and the mix of them are adopted. The maximum cross correlation method is chosen as the likelihood measurement. The disparity map of a resistance is gained by this method. To recover the 3D depth information, a stereo vision model so called Weak Disparity Vision Model (WDVM) is introduced. And the process of calibrating used calibrating template is listed. OpenGL is selected as the developing environment in the step of stereo display. The processing of 3D discrete data mainly includes the following steps: data pre-process, triangulation of 3D discrete points, triangular Bernstein-Bezier curve interpolation. Some works of 3D points pre-process and Delaunay triangulation (DT) of 2D discrete points are done in this paper. A new DT method is brought forward through the improvement of data structure and processing flow based on the algorithm of Cline-Renka.Several matching methods are used in the reconstruction experiments of a plastic templet and a resistance's surface. In the reconstruction experiment of micro channel's cross section, two methods were used: mixed characters and edge marks, and the reconstruction error of depth reaches 2.8% and 1.2% respectively. The result of experiments shows that it is possible to reconstruct and measure the microcosmic object through stereo matching method. But the matching elements and operators should be selected according to the characters of microcosmic objectand scene. Under the precondition of keeping enough information, the stereo matching algorithm should reduce the false matching rate and improve the computing speed to the best of its abilities.