基於互資訊之半全域匹配補償函數特性及作業參數擬定初探

Abstract

近年來利用影像產製高密度點雲為攝影測量和電腦視覺領域中之重要議題。基於互資訊之半全域匹配法利用多路徑進行匹配值加總，提升逐像元匹配運算效率。然而，為減少匹配錯誤，利用補償值於匹配值加總時進行平滑約制，因此路徑上的平滑約制補償值設定成為決定視差成果的關鍵。本研究分析補償值對半全域影像匹配之影響，並以補償值函數最小值設定模式進行補償函數特性分析，以視差影像之一致性像元數趨勢提供補償值設定參考，於實驗室影像和實際影像中探討作業參數擬定。

Over the last few years, dense image matching for point cloud generation has attracted research attention in both the photogrammetry and computer vision communities. The semi-global matching (SGM) algorithm based on mutual information is one of well-known methods which applies multi-directional smoothing constraint in cost aggregation to efficiently enhance the rate of stereo matching. On the other hand, the smoothing constraint, among others, is the core issue to equip the penalty function with sufficient power for reducing erroneous matches if appropriately chosen. For that, the purpose of this study is to characterize the smoothing constraint by setting mode of minimum of the penalty function. And offer reference of tuning parameters of penalty function by trend of consistency pixel numbers. Tests on Middlebury Stereo Datasets and real image have been carried out and evaluated.