The present invention relates to a method for improving the matching accuracy in a stereo matching based on the census transform by using an adaptive area weight value technique and a window having a loose shape. The method comprises: a first step which sets an adaptive area weight value window centered on a specific point of the left image, and an adaptive area weight value window, having an identical size with the same of the left image, centered on a point which is located within the maximum disparity prediction value and corresponds to the specific point of the left image in the right image; a second step which calculates an adaptive area weight value with respect to pixels in each weight value window of the left and the right images; a third step which sets a matching window with respect to pixels within the adaptive area weight value windows set in the left and the right images; a fourth step which calculates a raw matching cost with respect to pixels within the adaptive area weight value window by using the matching window; a fifth step which calculates the matching cost by using the calculated adaptive area weight value of pixels and the raw matching cost; and a sixth step which calculates a matching cost for other points located within the maximum disparity prediction value and corresponding to the specific point of the left image in the right image, and calculates a disparity value by means of comparison with the calculated matching cost.
展开▼