Research Of High Accuracy Contour Extraction Methods Based On ICT Images

With the increasing demand of high accuracy in industrial measurement and some related application, the subpixel localization technique has got great attention from technicians and is becoming a hot topic in image processing. It can increase the accuracy up to subpixel by software and break through the limits of classical edge detection algorithms whose accuracy is confined by the precision of imaging systems. Consequently, it is of great importance for proving more accurate data for later model reconstruction and virtual measurement without increasing the burden of hardware cost. The main research contents and contributions in the thesis are as follows:1. Discussion of subpixel edge detection algorithms. The algorithms based on surface fitting and those based on parameter model are discussed in this paper. Of the former class, a subvoxel edge detection algorithm based on 3-D Facet model is introduced, and to the problem of large computation, an improved acceleration method based on threshold segmentation is proposed. Of the latter one, we study the moment-based subpixel edge detection method, and aiming at the inapplicability of curve detection we introduce parabola model to moment-based curve detection and deduce the corresponding formulas which acquire edge parameters by solving a non-linear equations.2. Introduction of a region-based and supervision-based tracing strategy. In contour tracing of images with complex contour shape, the classical algorithms usually get into a trap or take false steps. To solve this problem, a region-based and supervision-based strategy is first introduced for contour tracing of image sequence in this paper. The region-based strategy employs different contour tracing algorithms in different regions according to shape and shape changes in the slice image, that is, in the regions with simple cross-sectional shape and great cross-sectional shape changes, the strategy adopts a classical tracing method, whereas in the regions with complex shape and little shape changes it uses a classical tracing algorithm combined with the supervision-based strategy which is directed by a given contour chain achieved from an adjacent slice image so that the detected contour chain acquires the same topology as the given one. The strategy improves tracing results significantly by combining with a classical algorithm.3. Experimental evaluation of subpixel edge detection algorithms. We develop programs of the subvoxel edge detection algorithm based on 3-D Facet model, the moment-based subpixel edge detection algorithm and the tracing method combined with the region-based and supervision-based strategy. In order to evaluate the accuracy, rudeness and influence of blurring an ideal line and circle generation method is introduced to produce an ideal image which is to be added with Gauss noises and Gaussian blurring respectively to provide three groups of simulation data for algorithm evaluation. Finally, experiments on a hollow turbine blade validate the feasibility of our methods.