Study Of Image Noise Reduction And Image Enhancement Based On PCNN

During the acquisition, storage and transmission of images, some of the problems, such as noises, low dynamic range, low contrast, and blurred edges usually occur in images because of imaging system, storage medium or environmental factors, which results in low quality of images and affects person's visual perception, so the recognition and understanding of image content is affected significantly. Hence, it is of the utmost importance to de-noise and enhance images in order to improve the quality of images.Image de-noising (or image filtering) is an image processing method which reduces or eliminates noises or the undesired signal in images on the premise of preserving the leading features such as details and edges as possible. The aim of image de-noising is that the de-noised image is as approximate to the uncontaminated or the "clean" one (that is, the reference image) as possible. That is the difference between the de-noised image and the reference one should be as small as possible. So, image de-noising uses the fidelity as its processing principle.Image enhancement is an application-oriented image processing method which emphasizes or highlights the information of interest, and suppresses or eliminates that of unimportant in images or which transforms images to make them more adaptable to analysis for man or machine. Image enhancement does not use the fidelity as a processing principle, and will not increase the amount of information, but highlights some features of interest, such as sharpening edges, improving contrast or stretching dynamic range, so that the recognition of image content is easier. That is, enhancement processing increases the intelligibility of images for man or machine. So, the enhanced images do not necessarily approximate to the original ones. The visual effect of images is of high subjectivity and has a close connection with human visual characteristics.This dissertation puts research emphasis on the properties of the three visual neural network models Pulse Coupled Neural Networks (PCNN), Intersecting Cortical Model (ICM) and Spiking Cortical Model (SCM), and the analysis of connection between the time matrixes of the visual neural networks and human visual characteristics, then applies the neural networks to image de-noising and image enhancement. The central content is abstracted as follows: 1. Analyze the auto wave transmission property of PCNN, study and discuss the corresponding relation between the linking matrix of PCNN and the structure element (SE) in mathematical morphology when PCNN is used in the simulation of binary erosion and dilation operation.The neuron of PCNN intercommunicates with its neighbors through the linking matrix. Morphological operations use a SE with a specialized shape as a "probe" to measure or extract the image geometrical characteristics to analyze image and recognize objects. So, in binary erosion and dilation operations, there is some kind of corresponding relation between them. Through extensive experiments, this relation is summarized simply. The insight into the relation is helpful to the comparison and full understanding of the image processing principle of PCNN and mathematical morphology.2. Combine PCNN and mathematical morphology to de-noise pulse noises in images.The pulse noise (salt and pepper noise) appears in images as isolated dark or light pixels. Moreover, it is some portion of pixels that are noisy. So to preserve the information carried by the uncorrupted pixels noise removal as well as the locating of detecting of noisy pixels is of importance. The synchronizing pulse burst property of PCNN is used to detect the isolated noisy pixels and then the mathematical morphology operation is used to remove the detected noisy pixels. This method can not only save time but also decrease the blurring of details and textures. In the detection of noisy pixels, ICM and SCM are also used instead of PCNN.3. Study the relation between visual neural network and human visual characteristics, taking SCM for an example.PCNN, ICM and SCM have a biological background and have similar properties to human visual characteristics to a certain degree. Mathematical analysis shows that the time matrix of SCM has the same expression with Weber-Fechner law except parameters. Experiments show that the negative time matrix of an image is indeed a contrast-stretched version of the input except that there is some gray scale loss because of different level of details processing in dark region and light region for the time matrix. Combining the time matrix with other methods, such as Unsharp Masking or histogram modification can overcome the problem and enhance images more effectively.