Image Processing Based On Pulse Coupled Neuron Networks And Its Implementation On DSP

Pulse Coupled Neural Network (PCNN) is a novel artificial neural network which comes from the research of mammal's visual properties and has a biology background. As PCNN is much similar to the process course of vision system, its working mechanism and applications on image processing have received wide attention. So far, PCNN has been used in kinds of applications of digital image processing. The paper mainly studies the basic characteristics of PCNN and makes a further exploration of its applications in image processing. In this paper, the following works have been done according to the latest research of PCNN:1. PCNN differs from conventional artificial neural networks by having the significant autowave characteristic. So after analyzing the PCNN model and its working mechanism, we study the autowave characteristic of PCNN. Actually, when change the PCNN parameters, it will result in the alteration of the wave behaviors. Then we present a novel method for blood cell image segmentation and counting based on PCNN autowave.2. In order to overcome the weak adaptability and the difficulty in selecting the adaptive threshold of the traditional image binaryzation and to improve the reliability lacking in the traditional single evaluation of image segmentation, an image binaryzation method based on PCNN is investigated, and the corresponding parameters are selected. Afterwards, a composite segmentation evaluation method comprehensively considering various evaluation criteria is proposed. Experimental results show that the PCNN-based image binaryzation method is of high accuracy and is suitable for the segmentation of varied images, and that, as compared with the traditional single evaluation methods, the proposed composite method can evaluates the performances of segmentation algorithms more objectively and accurately.3. Studied the human vision system psychophysics related to image processing, and give a brief introduction to the second generation image compression coding based on HVS. Then elaborate on the irregular segmented regions coding (ISRC), and give its coding and decoding algorithm framework, includes image contour coding, Schmidt orthogonal basis reconstruction of segmented regions. We also proposed a fast algorithm for connected components labeling.4. Combined the PCNN based segmentation with ISRC, the paper studied the PCNN based irregular segmented regions coding (PCNNISRC). Experimental results indicate that PCNNISRC can not only get a high compression ratio, but also protect image details well. Compared with traditional block transform coding (BTC), PCNNISRC method has an advantage over BTC both on compression ratio and reconstructed image quality. At last we implement the PCNNISRC algorithm on the TMS320C6713 DSK board.