The Research Of Improved Pulse-Coupled Neural Networks Applied In Image Segmentation

Pulse-Coupled Neural Network, which is based on mammalian brain’s visual cortex neuron model, is summarized by Eckhorn and other scientists in the case of the stimulation which can cause the neuron distributing pulse sequence and synchronous oscillation. This model is based on the background in biology, which is similar to the biological visual system, and has the ability to process images. So it is very suitable for the application in the field of digital image processing. In view of the image segmentation problem, a simplified and optimized model is proposed, and the new model is applied to the medical red blood cells detection and counting problem. The main work done are expounded as follows:i) In order to make it more applicable in solving the problem of digital image segmentation, the Pulse-Coupled Neural Network model is optimized, which mainly reflected in the following two aspects:a. In view of the fact that the parameters in this model play different effect in image segmentation problem, the input part of model is simplified which mainly reflected in two aspects. The input of neuron’s F channel only considers the influence of external image pixel’s gray value, the input of neuron’s L channel only considers the influence of eight corresponded neighborhood neurons’output in last moment.b. In view of the problem that the parameters of PCNN model need lots of artificial test to determine, the Genetic Algorithm is adopted to optimize. Having the ability which can acquire and accumulate the knowledge of the search space automatically, Genetic Algorithm make the key parameters in the Pulse-Coupled Neural Network set value adaptively.ii) With the development of the large scale integrated circuit technology, algorithm implemented by hardware method has become a trend. In order to keep up with the trend, the improved algorithm is implemented by FPGA hardware developing board. And the following three aspects are solved during the implementation.a. In view of the storage resources limitation of the FPGA development board and design cost constraint, the algorithm is improved, which makes the number of neurons reduced to only one, and adopts serial method to segment the whole image.b. For ease of the hardware implementation, the binary encoding method, the random league selection operator, the multi-point crossover operator and the basic position mutation operator are chosen in the Genetic Algorithm design.c. In order to achieve the display of the image after segmentation, an Image-Display module is designed in this article, which exports data signals and controls signals to VGA interface to display the segmented image intuitively on a computer screen.ⅲ) In this paper, the improved algorithm is applied in automatic counting of the medical red blood cell images, which solved the problem using common segmentation algorithms cannot achieve the ideal effect, and the result proves the superiority of the improved algorithm.