Determining the window size based on modified 45◦ rotation skeleton-based region method

In local methods for the matching of stereo images, determine the optimal window size plays an important role in accurate estimation of disparity map. To achieve the optimum window size we will use the modified 45◦ rotation skeleton-based region method (RSBR). Pixels within a similar range in terms of color and levels intensity probably have the same disparity maps. In this paper, by using color image segmentation algorithm, we obtain the initial window of all objects in the image at first; after that we obtain the vertical arms of each window and rotate them 45○ counter clockwise. Finally all the pixels in the same level are judged simultaneously, to determine the similarity between the central pixel and the neighbor pixels. The law of similarity is implemented at two stages such that at each stage the including pixels in the stage are used to determine the including pixels for the next stage. This method is called the Modified Rotation-Skeleton-Based Region (MRSBR). The proposed method has been able to provide more optimal solution than other similar methods based on two aspects: First, it identifies the noise in each level and passes through them; second, for each pixel in the coordinates of the image in which brightness changes have a bad effect on the image, it can optimally determine the window size.

Keywords:

depth extraction , stereo images matching , window size determination