Corners in an image were detected by multi-scale Harris operator,and they were taken as initial interest points. Since adaptive non-maximal suppression eliminated lots of potential matching points,condition theory was applied to control the number of initial interest points. The bad conditioned points were eliminated,so the computational complexity of the following process was decreased and the efficiency of the algorithm was improved. Meanwhile,matching points were mostly kept. Features under the control of condition theory were called well conditioned features. Since the location of initial Harris corner had offset and false corners were produced,sub-pixel localization technique was used to determine the location of corners,and the false and unstable corners were eliminated in this process. PCA-SIFT was applied to describe the feature points and their neighbors to get vectors. Finally,Euclidean distance between vectors was used to determine whether two points match or not. Experimental results show that the efficiency is improved significantly,the proposed algorithm is good at feature matching,and has good robustness to geometrical transformations,image noise and illumination change.