Integration of focus and defocus analysis with color stereo for three-dimensional shape recovery

Recovering the 3D (three-dimensional) shape of objects in a scene is an important topic in computer vision. Several methods including passive and active methods have been developed by researchers to investigate this topic. Two kinds of information are recovered during the 3D shape recovery. One is the geometric information about the shape of the object. The other is the photometric information about the light energy reflected by the object surface. One effective method is developed in this research and its algorithms and results are presented in this dissertation.; The method developed in this research integrates three important techniques in machine vision---IFA (Image Focus Analysis), IDA (Image Defocus Analysis), and SIA (Stereo Image Analysis). IDA is used to obtain a rough shape of the object and the spatial range occupied by the object. It uses less image data and less computation than IFA. IFA uses the results of IDA to reduce the image acquisition data and refines the estimated shape of the object. IFA also recovers the focused image of the object. The refined shape obtained by IFA is used by SIA to limit the search range for finding conjugate points as well as to reduce the correspondence problem associated with SIA. SIA then recovers the 3D shape with better accuracy. This integrated method results in optimal performance with respect to image data, computational cost, and accuracy in the recovered shape of the object.; Color information is used and other techniques including multiple baseline stereo and multiple resolution stereo are integrated together to increase the accuracy and speed. IDA and IFA are incorporated with SIA to deal with the correspondence and occlusion problems. Results are improved and are further improved by using blurred images for stereo matching. Calibrations and experiments are done on a practical digital stereo vision system. A 360-degree 3D model with both geometric and photometric information is recovered by integrating multiple partial 3D models of an object recovered by the integrated method developed in this research. Results are presented and the effectiveness and usefulness are validated.