CAJ | 학술논문

针对尺度不变特征变换( SIFT)匹配中存在的误匹配问题和立体图像特点，提出一种误匹配消除方法。对体视显微图像进行SIFT特征匹配初步得到匹配对，结合体视显微镜标定参数，计算三维点云坐标。将三维点云分别投影到左、右图像中得到新的匹配对，新投影点的图像坐标分别与原来匹配点的图像坐标相减，生成投影向量集。通过左、右2个投影向量集幅值和方向的异常值剔除，实现误匹配消除。实验结果表明，实验图像的误匹配消除率达到100%，同时不消除正确匹配点，提高了匹配精度。
침대척도불변특정변환( SIFT)필배중존재적오필배문제화입체도상특점，제출일충오필배소제방법。대체시현미도상진행SIFT특정필배초보득도필배대，결합체시현미경표정삼수，계산삼유점운좌표。장삼유점운분별투영도좌、우도상중득도신적필배대，신투영점적도상좌표분별여원래필배점적도상좌표상감，생성투영향량집。통과좌、우2개투영향량집폭치화방향적이상치척제，실현오필배소제。실험결과표명，실험도상적오필배소제솔체도100%，동시불소제정학필배점，제고료필배정도。
In order to overcome the difficulty of eliminating stereo mismatching pairs on Scale Invariant Feature Transform( SIFT) feature matching,an algorithm aiming to remove error match pairs of Stereo Light Microscope( SLM) images is proposed. SIFT feature matching pairs are roughly found,and then calculate 3D coordinate of point cloud using SLM calibration parameters. New match pairs can be obtained by projecting the point cloud to left, right images and generate projection vector set by subtracting image coordinate of new projection point from one of original point, respectively. Outliers of magnitude and orientation in left,right projection vector sets are eliminated to remove error match pairs. Experimental result presents that the error pairs removal percentage of experimental images in proposed algorithm reaches 100%,meanwhile it does not eliminate correct matching pairs,improving the precision of stereo match.