基于方向梯度极值的手形轮廓跟踪算法苑玮琦.pdf

18 7 Vol. 18 No. 7 2010 7 Optics and Precision Engineering Jul. 2010 1004-924X( 2010) 07- 1 75-09 苑玮琦, 董 茜, 桑海峰 ( 沈阳工业大学视觉检测技术研究所, 辽宁沈阳110870) : , , , , , , ( H KUST ) , , 100% , 85. 8% , 99.4% , , : 手形; 边缘检测; 轮廓跟踪; 灰度梯度 :T P391. 41 :A doi: 10. 3788/ OPE. 1 75 Hand shape contour tracking method based on directional gradient extremum YUAN We-i qi, DONG Qian, SANG Ha-i feng (Comp uter Vision Group , Shenyang Univ ersity of Technology , Shenyang 110870, China) Abstract: Hand contours are hard to be extracted correctly w hen hand images are illuminated by a un- symmetrical light. T herefore, this paper proposes a contour tracking algorithm based on the directional maximal gradient value according to the characteristic that values of edge pixels turn sharply in the vertical direction of an edge. The algorithm firstly finds the starting point of hand contour, then de- pending on some searching rules, it calculates the gradient values of candidate points in a local region, chooses the point w hose gradient value is maximal in the candidate set of local region and tracks it point by point to get the final contour. T racking experiments are carried out on both the hand image database formed by ourselves and the HandImage database from Hong Kong U niversity of Science and Technology( HKU ST) . Experimental results indicate that the accuracy of contour tracking is 100% in ourselves. database, and 85. 8% in the database from HKUST . Moreover, the accuracy of contour tracking in the images eligible to our refined condition from HKUST database is 99. 4%. T hese data show that the algorithm can directly track out the accurate, consecutive and integral hand contour in