Applications of Pulse-Coupled Neural Networks

The image captured by eyes is transmitted to brain by the optic nerve, and the image signal is transferred in the fiber pathways and finally processed by the primate visual cortex dominantly. The primate visual cortex is devoted to visual processing, and nearly all visual signals reach the cortex via the primary visual cortex. The primary visual cortex is the largest and most important visual cortical area, and does so when neurons in the cortex fire action potentials as stimuli appear within their receptive fields. Signal produced in neurons is transferred to their neighbors by means of localized contact of synapses, which are located on the dendrites and also on the neuron cell body. Electrical charges produced at the synapses propagate to the soma and produce a net postsynaptic potential. If the postsynaptic potential is large enough to exceed a threshold value, the neuron generates an action potential. Synchronized Gama oscillations (30 – 100 Hz) were found in the primary visual cortex of mammalian [1], [2]. In Ref. [2], the linking field model was proposed based on the hypothesis that neuronal pulse synchronizations can be partitioned into two types: stimulus-forced and stimulus-induced synchronizations. Stimulus-forced synchronizations are directly driven by stimulus transients and establish fast but crude sketches of association in the visual cortex, while stimulus-induced synchronizations are believed to be produced via process among local neural oscillations that are mutually connected. The feeding and the linking create the membrane potential.

Image filtering is able to enhance (or otherwise modify, warp, and mutilate) images and create a new image as a result of processing the pixels of an existing image. Each of pixels in the output image is computed as a function of one or several pixels in the input image, usually located near the output pixel. Different kinds of functions produce different results, and are usually used to remove different noise.

Image segmentation is a partitioning of an image into constituent parts according to its attributes such as pixel intensity, spectral values, and/or textural properties [1]. Image binarization segmentation which is defined as dividing an image into objects and background is the most fundamental and important processing step and common, basic and key technique in the research of object identification, image understanding and computer vision. The performance of image segmentation will impact directly on the subsequent object identification and image understanding. There are many methods of image segmentation and the simplest and most effective one is the method based on the gray-level threshold, but it is very difficult to select a appropriate threshold. In this chapter, two image segmentation approaches with PCNN: one based on entropy or cross-entropy and the other based on genetic algorithm (GA), are introduced.

The large amount of data bring difficulties for storage and transmission of digital images. For instance, a typical uncompressed scanned image of size 2 480 × 3 500 will take up approximately 25 megabytes of storage space. The data compression techniques have been used to reduce the amount of data required by representing an image with no or little distortion as far as possible. It means that the digital data contain much redundancy, which is of no or little use for image representation. The aim of information coding is to represent the effective information accurately with less code.

Image enhancement is one of the most commonly used methods in the image pre-processing aiming at improving visual effects for a specific purpose. In other words, the region of interest of images will be prominent after the image enhancement for easier analysis by human or computers. However, there exists no general or uniform standard for evaluating the enhancement quality objectively, because image enhancement methods usually depend on the special needs for some particular applications. In most cases, the enhancement effects are evaluated by the visual perception.

The PCNN is a biologically inspired neural network, widely applied in the field of image processing such as the image segmentation, image enhancement, and pattern recognition. This chapter will describe its application in image fusion. Firstly, it will briefly review the development of the image fusion based on the PCNN, and then simply introduce some medical image fusion methods [1]. Finally, it will describe a new method of the multi-focus image fusion suggested in Ref. [2].

In Ref.[1], Johnson et al. proved that the output pulse images of PCNN could be converted to a rotation, scale, distortion, and intensity invariant time series which is the summation of 1’s in the binary pulse images. Refs. [2], [3] introduced some more efficient statistical measures than time series, and used transform to characterize pulse images. This chapter will introduce the application of PCNN in the field of pattern recognition, especially, object recognition with feature extraction based on the PCNN. It is organized below. Section 7.1 briefly reviews some methods of feature extraction based on PCNN, including time series, energy time series, entropy series, energy entropy, average residual and standard deviation. Section 7.2 investigates the anti-noise PCNN feature extraction in the noised image recognition [4]. Section 7.3 describes a feature extraction method combining PCNN with histogram vector barycenter of images [5]. Section 7.4 presents some applications of feature extraction in content-based texture retrieval [3]. Section 7.5 presents its application in the iris recognition system [6], and Section 7.6 is the summary.

Combinatorial optimization is an archaic problem that brings people puzzles until now. This kind of problems are, given the restricted conditions, to consider all risk synthetically, and find the variable that makes the objective function greatest or least. Graph Theory is always used as the mathematical basis for solving this problem. More and more people devote themselves into this area, including many famous genius and amateurs. A classical example is the seven bridges problem [1].

The PCNN image processing algorithms are generally programmed on PC platform [1], [2]. These algorithms have fully demonstrated their outstanding performance. With the development of large-scale integrated circuit technology, the hardware implementation of neural network becomes more and more imperative. The combination of the DSP, FPGA (Field-Programmable Gate Array) and other hardware with neural network provides a standout platform for further research and application of neural network information processing. This chapter will discuss the FPGA implementation of the PCNN image processing algorithm.