Scene discrimination by recalling with visual neural system

In this paper, a neural system based on human visual perceptive model for scene discrimination is proposed. The scenery represented by color image is memorized by the neural network based system as perceptually simplified scene image. A blurred and noisy uncertain scene image is recognized as the original image by recurrent processing with parallel Hopfield-type neural networks. In order to reduce color information naturally, quantization and segmentation in L*a*b space is executed in the preceding step. Several input images such as slightly shifted, noisy, partial or mixed scenes are used in the discrimination. It is shown that the blurred images are effectively discriminated by recalling process with the proposed visual neural system. Effectiveness of quantized segmentation for original color scene images is also examined in the simulations.