Dissertation
Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

The Applications in the Medical Image Processing Based on Pulse Coupled Neural Network

Author FanHongBin
Tutor ZhangXianQuan
School Guangxi Normal University
Course Applied Computer Technology
Keywords Medical image Pulse Coupled Neural Networks Mixed noise filtering Image Segmentation Image Edge Detection
CLC TP391.41
Type Master's thesis
Year 2009
Downloads 60
Quotes 2
Download Dissertation

The development of one of the fastest growing technology areas of medical imaging has become widely used in diagnosis and treatment, is an important means and tools in modern medicine. Medical image processing is a very important aspect of medical imaging technology, can effectively medical image processing, thereby improving the utilization of medical image information, to help extract lesion feature information, clinicians observed lesions more direct and specific, to improve the diagnosis rate. Pulse coupled neural network as the third generation of artificial neural networks, has a biology background is created by simulated mammalian visual cortex nerve cell activity and neural network model is simplified approximation of real neurons. This model is similar to the two-dimensional space of image grayscale similar pixel characteristics of the packet, and can reduce the image local gray difference, make up the image of local small intermittent characteristics. Link domain characteristics and dynamic threshold attenuation characteristics similar state neurons sync output pulse. Therefore, it is closer to visual system processing images, especially its the nonlinear modulation characteristics in medical image processing has a broad application prospects. Improved traditional pulse coupled neural network model theory and its application to medical image processing, the main job of the work of the following aspects: 1, complex medical image imaging system acquisition, display and transmission process, inevitably introduced a variety of noise is a Gaussian noise and impulse noise. Similarity clusters, according to the nature and characteristics of impulse noise, Gaussian noise pulse coupled neural network coupling link state grant synchronization pulse has spatial proximity, the brightness intensity similar to the input neurons will fire at the same time. The most pixel brightness mean pulse coupled neural network sync pulse payment feature to locate the position of the point of impulse noise and Gaussian noise, the only noise pixel processing in pulse coupled neural network control the brightness of the neighborhood is basically the same as the noise pixel the gray value, to improve the performance of the image filtering. Non-noise pixel, before and after filtering of the pixel gray value does not change, thus ensuring that the noise pollution of the image before and after filtering image without any distortion, and distortion on a point of non-noise pixel. Algorithm, image adaptive filtering, compared with the traditional pulse coupled neural network, without the need for multiple traversal, improve efficiency and save time. Well this method in medical image mixed noise removal retained the image detail and edge information, the treatment effect was significantly better than the mean filter, median filtering, Wiener filtering denoising methods, especially for high-density pulse noise, high variance Gaussian noise and mixed noise medical image processing. 2, Medical Image Segmentation Medical Image Understanding, the basis of the three-dimensional reconstruction, visualization, registration processing, has important clinical significance in the diagnosis and treatment of the disease. This article will pulse coupled neural network is introduced to the field of medical image segmentation, a medical image segmentation based on the units connected pulse coupled neural networks, maximum cross entropy and sub mean filtering. The algorithm used units connected pulse coupled neural network levels along the direction of the high-brightness values ??followed by decomposition of grayscale images using the the image maximum cross entropy the units connected pulse coupled neural network for image segmentation to determine optimal results, while the sub-mean filtering to overcome the impact of noise on the segmentation process. The algorithm to use units connected pulse coupled neural network to solve the traditional pulse coupled neural network image segmentation parameter selection problems, optimization of the pulse coupled neural network model unify nerve yuan link input channel signal, the dynamic threshold division, region of space location relations and image cross entropy organically combine natural image and automatic segmentation using pulse coupled neural network unit connected to the pulse propagation, to solve the problem of unable to determine the number of iterations. Proposed algorithm can automatically split a variety of medical images, medical images can be divided into a large number of details, detected lesions, certain anti-noise capability, faster processing, segmentation is better than the threshold segmentation, Otsu adaptive thresholding segmentation and other methods. Edge detection in medical image, this paper presents a based on The units connected pulse coupled neural network medical image edge detection method, the first unit connected pulse coupled neural network for medical image segmentation as a binary image, and then lit area (background natural and complete) and dark areas (target) of the ignition binary image exclusive or firing pulse in accordance with the shape of the object naturally synchronous communication, in order to quickly obtain Target edge. The algorithm does not need to select the parameters, and neuron parameters on the results are not sensitive cluster mechanism of the network of neighborhood similarity detection algorithm has a certain degree of self-adaptive, can quickly and naturally and accurately extract the edge of the gray image. Compared with the traditional image edge detection algorithm, the proposed algorithm reduce edge undetected retained a richer details, effectively guarantee the continuity, integrity and precise positioning of the edge of the medical image, has strong self-adaptability.

Related Dissertations
More Dissertations