CAJ | 학술논문

通过研究Harris角点检测算法时发现，虽然该算法对棋盘角点有很好的检测能力，但对于一些立体物角点提取时，在物体边线上存在较多伪角点，同时也有抗噪能力差和运算速度慢的问题。针对以上情况，提出一种Harris邻域像素取差法与SUSAN相结合的改进算法。首先，运用邻域像素相似法，计算当前像素点8邻域内灰度值与其核点相似的像素点个数，初步得出候选角点集；然后结合SUSAN算法思想，进一步排除候选角点集内的伪角点。实验结果表明该方法能提高角点检测的准确度。
통과연구Harris각점검측산법시발현，수연해산법대기반각점유흔호적검측능력，단대우일사입체물각점제취시，재물체변선상존재교다위각점，동시야유항조능력차화운산속도만적문제。침대이상정황，제출일충Harris린역상소취차법여SUSAN상결합적개진산법。수선，운용린역상소상사법，계산당전상소점8린역내회도치여기핵점상사적상소점개수，초보득출후선각점집；연후결합SUSAN산법사상，진일보배제후선각점집내적위각점。실험결과표명해방법능제고각점검측적준학도。
Through studying Harris corner detection algorithm we find that though the algorithm has good detection capability on chessboard corner, but when extracting some cubic corners there are quite a few false corners existing on the edge lines of the object , and meanwhile it also has the problems of sensitive to noises and low computation speed .In light of the situation above , in this paper we present an improved algorithm, which combines the Harris neighbourhood pixels offset extraction method with SUSAN algorithm .First, the neighbourhood pixels similarity method is applied to calculate the number of pixel points within the 8 neighbourhood of current pixel which gray values are similar to their kernel points , and initially gets the candidate corner set;then in combination with SUSAN algorithm idea , it further removes the false corners in candidate corner set .Experimental results show that this method can improve the accuracy of corner detection .