A kind of feature point matching algorithm based on the points feature and random sampling consensus is proposed. The describing function of feature points with simple structure is defined through simulating the Rank Order coding in the artificial neural network. A novel geometric model named the statistical deflection angle (SDA) is proposed to describe the points in the image. The SDA describes a feature point by combing with the local structure around the point and the global statistical information of all the points in the image. Then a point matching strategy improving random sample consensus is used to remove the outer points, the matching function with higher accuracy is obtained through iteration. It is proved by the experiment results that better effects in matching accuracy can be obtained by the method.