Abstract: False corners and inaccurate orientation in key point extraction algorithms will result in low matching rate and low image registration accuracy and speed. A new method is proposed to solve the problem. A set of strongest interest points is worked out in the algorithm to eliminate false corners. Interest points in the set have high characteristic strength and meet several criteria. The same coordinate is applied to ensure orientation accuracy. With similarity transformation， brightness change and noise， a series of images are tested with the new algorithm proposed and traditional algorithms， respectively. The results show that the new algorithm has better adaptability and anti-noise performance， and is more effective in feature points extraction.

Key words: key point extraction, point matching, voting strategy, image registration, false corner