Binary Images Classification Algorithm Based on Moore-Penrose Inverse Matrix

摘要

This paper presents a novel binary images classification algorithm that uses Moore-Penrose inverse matrix to update a weight matrix based on the error obtained for an image sample in a training set. It can be seen as a perceptron in terms of matrices. Weight matrix update is made estimating a desire weight for a desire output using Moore-Penrose inverse. Evaluation was made over a 3 image classification problem example and using a sign traffic dataset. Obtained results shows good algorithm performance and accuracy if image size is appropriate. Results suggest training images must have rich features and discriminant information between classes.