论文部分内容阅读
A new stereo matching algorithm ispresented. Firstly, the bicubic B-spline smoothing fil-ter operator is applied to the initial image pair toextract edges. Then area-based matching which em-ploys the modified SSD (sum of squared differences)algorithm is applied to the two binary edge imagesto obtain a coarse level matching. A template win-dow with slightly larger size is adopted to achieve arobust and reliable result in this process. Finally, in-tensity information is added to obtain matching on afiner level. The use of edge images simplifies the com-putation and the coarse-to-fine matching procedure isconsistent with the formation process of human vi-sion. The cost function is modified according to thestatistical model of the disparity distribution withinthe window, which increases the accuracy of the algo-rithm. The method has been tested with real stereoimages and the results demonstrate its accuracy andstability.