CAJ | 학술논문

图割算法是图像分割方法中的一种高效的最优化计算方法,针对图像中目标物体的旋转尺度光照变化导致的分割不准确问题,提出了一种基于SIFT(Scale-lnvariant Feature Transform)特征的图割(Graph Cuts)算法;该方法将SIFT特征的尺度旋转不变性和图割算法的准确快速性结合在一起,通过提取图像中物体SIFT特征点做为图割算法的种子点,求解最小能量函数快速从而获得该图像的最优分割;实验结果表明,该方法鲁棒性较好,能准确地分割出目标物体在图像中的轮廓.
도할산법시도상분할방법중적일충고효적최우화계산방법,침대도상중목표물체적선전척도광조변화도치적분할불준학문제,제출료일충기우SIFT(Scale-lnvariant Feature Transform)특정적도할(Graph Cuts)산법;해방법장SIFT특정적척도선전불변성화도할산법적준학쾌속성결합재일기,통과제취도상중물체SIFT특정점주위도할산법적충자점,구해최소능량함수쾌속종이획득해도상적최우분할;실험결과표명,해방법로봉성교호,능준학지분할출목표물체재도상중적륜곽.
The graph cuts algorithm is one of high effective optimal methods in the Image segmentations. To improve the accuracy of segmentation when object changes rotations, scale and illumination, a method which combines Scale - Invariant Feature Transform (SIFT) with graph cuts is proposed. The existence of object is recognized by voting processing of SIFT key points and the object region is cut out by graph cut using SIFT key points as seeds by max-flow/min -flow algorithm. Experimental results show this method has a high accuracy and strong robustness.