Stereo computation using radial adaptive windows

While stereo algorithms using global methods produce good depth maps, local methods are still preferred for their efficiency. We present an algorithm that determines the adaptive support windows by radial computations. Motivated by the experiences of human vision, we use the certainties of the initial disparity distributions to determine the support window. Certain other cues, such as differences of the initial disparity distributions, and color similarities, are integrated to determine the weights in support window. Extensive experiments have been conducted. The algorithm performs well in textureless areas, and preserving sharp disparity discontinuities.