Abstract: To extract efficient image features is an important task for image recognition. In this paper, we first derive a new algebraic concept: similarity of matrices, establish a similar discrimination function of images, and further present a feature extraction and recognition method based on the similar discrimination function. Finally, this method is used to solve the recognition problem of human face images which is one of the most difficult problems in pattern recognition. Experimental results have shown that our feature extraction and recognition method is very effective, and the designed classifier has high recognition performance.