An Experimental Study on the Auto-association Neural Networks with Electronic Image Processing

Based on the mechanism of auto-association, this paper proposed a new type of auto-association neural networks which can classify images and restore distorted images. Meticulous experiments were designed with Matlab language to test the networks’ performance on image recognition and image restoration. In addition to random noise test, occlusion noises were included in the experiments too. The comparison of different number of recurrent layers settings in neural networks was discussed in details.