Abstract
One of the most significant and classic problems on area-based stereo vision is mismatching caused by photometric variations in the stereo images. Some approaches to the problem have applied spatial filters to images to decrease the variations in success to some extent with special scenes. These methods, however, could not solve an issue where local photometric variations are on objects with various directed surfaces (e.g., a fish scale). Accordingly, we propose the area-based stereo correspondence method with the spatial filter named “Standardized Coefficient Filter” (SC Filter) for that issue, and analyze the filter qualitatively. SC Filter is a kind of high-emphasis filter based on statistical analysis. This filter has two characteristics in amplifying high frequency components; 1) in direct proportional to mean curvature of low frequency components in a window, 2) in inversely proportional to absolute value of mean gradient of low frequency components in a window. In this paper, we show that SC Filter keeps stable output in the case of wave length of low frequency components varied, notwithstanding that LoG and zero-mean filter outputs varied. And we describe that the area-based stereo correspondence method with SC Filter is able to solve the issue where local photometric variations are on objects with various directed surfaces, too.
