Abstract
Shape From Focus/Defocus (SFF/SFD) has been known as a depth estimation technique by using focus information. The technique estimates the depth for each point of a target object based on a focus measure extracted for each block on a set of multi-focus images. The authors have proposed a 3-D whole shape reconstruction technique, which first obtains a set of multi-view and multi-focus images for a target object and then integrates the shape models obtained from each viewpoint with the SFF/SFD. In this technique, the depth estimation error is expected to be reduced by using not only a single set of multi-focus images from a single viewpoint but also another set from another viewpoint together. Although we have realized such an idea in our previous report, the criteria for estimating depth values was not necessary optimal. This report thus introduces a new criteria which uses the derivative of an approximating function for the set of the focus measures, and gives some experimental results to show its effectiveness.
