Adaptive Support-Weight (ASW)是一種以視窗為基礎的立體匹配技術,不同於使用全域最佳化得到深度圖的方法,ASW對每個像素位置獨立進行運算,可以用較低的複雜度得到不錯的效果。視窗大小對於ASW的準確度有很大的影響,為了得到較準確的深度圖,通常需要使用很大的視窗,但這對ASW的運算複雜度造成很大的負擔。本論文提出變動視窗大小以降低ASW運算複雜度的方法,我們先使用較小視窗,再根據所得到的視差結果進行分析。與鄰近像素視差相同者歸類為可信的位置,並且停止搜尋:與鄰近像素視差不同者先歸類為不可信的位置,並對這些位置重新使用較大的視窗;反覆以上過程直到最大預設視窗(即原始ASW視窗)為止。本論文使用Middlebury網站上的標準測試影像來進行實驗,實驗結果顯示我們提出的方法其深度圖的品質與ASW幾乎相同,而運算複雜度則降低20%以上。
Adaptive Support-Weight (ASW) is a window-based stereo-matching technique. Unlike the complicated global optimization methods, ASW calculates the disparity independently on each pixel. ASW achieves good results with less computational complexity. However, ASW is sensitive to the window size. A larger window is needed for better accuracy, which in turn causes the computational burden. In this thesis, variable window-size is proposed for ASW to reduce the computation. The disparity obtained from a smaller window-size is first analyzed. Those pixels with the same disparity values with adjacent pixels are regarded confident and cease further search. The other pixel locations are regarded unconfident, and the disparity values are re-calculated with a larger window. The above procedure iterates until the window size reaches its largest value (for the original ASW). The proposed method is evaluated using the standard test images in the Middlebury website. Experimental results show that the proposed method produces almost identical depth maps as the original ASW while the computational complexity is reduced by at least 20%.