2010 A biologically inspired neural network for image enhancement

ቤተ መጻሕፍቲ ባይዱ
Where, A is the decay rate constant; D is the baseline activities for the cell; E , F are nonnegative constants representing the upper and lower bounds of the neural
!!...Xk (i,j) -A[Xk (i,j)-D]+ [ E-Xk (i,j)]Ck (i,j) dt -[ F +Xk (i,j)]Sk (i,j)
=
(1)
This work is supported by grants from the National Basic Research (973) Program of China (NO. 2007CB714406), aud supported by the Fuudamental Research Fuuds for the Centtal Universities (NO. ZYGX2009XOO3, ZYGX2009Z005), aud supported by a grant from the Key Program of the Science Fouudation for Young Scholars in University of Electronic Science aud Technology of China (NO. JX0804).
978-1-4244-7371-7/lO/$26.00©20lO IEEE
2010 International Symposium on Intelligent Signal Processing and Communication Systems (lSPACS 2010) December 6-8, 2010
Yinghua Li, Tian Pu, Jian Cheng
School of Electronic Engineering, University of Electronic Science and Technology of China, 611731, liyinghua666 @, courierpt @, pami.cheng @
201O International Symposium on Intelligent Signal Processing and Communication Systems (lSPACS 20lO) December 6-8, 20lO
A BIOLOGICALLY INSPIRED NEURAL NETWORK FOR IMAGE ENHANCEMENT *
2. COLOR IMAGE ENHANCEMENT BASED ON IMPROVED PCNN 2.1. The Property of Visual Confrontation
network;
Image enhancement is of great importance in many applications, due to some features are hardly detected by eyes in an image. Since the images are observed by human, it is reasonable to develop the psychophysically-derived methods. Therefore, fmding a suitable visual model is a promising way to develop new image processing methods. There are a wealth of published literatures modeling the human visual system [1]�[2]. Here, we pay attention to the pulse coupled neural network (peNN) and the feed­ forward ON-OFF shunting neural network. The pulse coupled neural network (peNN) is a novel neural network different from the traditional ones, which was developed by Eckhom and his colleagues, based on the experimental observations of synchronous pulse bursts in mammal visual cortex[3]. It has significant potential applications in
ABSTRACT
Chengdu, Sichuan, China
A promising trend of image processing is to incorporate some knowledge on human visual system. In this paper, we propose an improved pulse coupled neural network (peNN) for image enhancement. We apply the passive membrane equation, which is known as a model for describing the ON-OFF opponent property of the receptive fields of the retinal ganglion cells, as the linking field to modulate feeding field input of the peNN and obtain the enhanced neural pulse as the output image. Initially, the RGB image is converted to luminance and chrominance images. Only the achromatic image is enhanced. Finally the RGB image is reconstructed from the enhanced luminance component along with the original chrominance component. The experimental results show the effectiveness of the method.
*
Physiological studies show that the receptive fields of ganglion cells in retina organize in two antagonistic center­ surround forms: excitatory ON-center with inhibitory OFF­ surround cells and inhibitory OFF-center with excitatory ON-surround cells[7]. The opponent mechanism in receptive fields results in that ganglion cells primarily transfer contrasts rather than light intensities. The passive membrane equation (PME), initially founded by Hodgkin and Huxley as a model, is expanded by Grossberg[8] to model the neural dynamics of the ON­ OFF receptive fields. It is described as the following equation:
activity, respectively; G{i,j)are inputs in the center neuron; 5;,(i,j) are inputs in the surround neurons. Studies show that the PME provides outputs to compute the ratio contrast in the image. We use the PME as a contrast extractor in the processing workflow.
Index Terms-pulse coupled neural opponent neural network; image enhancement;
1. INTRODUCTION
image processing and has been widely applied to image fusion[4], segmentation[5] and image denoising[6]. The ON-OFF shunting neural network is a model to understand the real-time adaptive response of the receptive fields of the retinal ganglion cells for complex and dynamic afferent stimuli. Knowing this, our motivation of incorporating these two neural networks is that it is more analogous to the retinal processing of the human visual system. As mentioned above, in this paper, we present a method based on an improved peNN. We use the response of the ON-OFF shunting neural network[7] as the linking fields of the peNN, thus the enhanced contrast components are combined with input signals through the state dependant the modulation of the peNN. Applications of this algorithm on various color images yield promising results.