Key Technologies Of Image Processing Based On Intersecting Cortical Model

The research topic of this dissertation is Key Technologies of Image Processing based on Intersecting Cortical Model. Intersecting Cortical Model (ICM) is a single layer neural network. ICM is based on the Eckhorn's anatomic research result of the cat vision cortex in the 1970s'. Based on the integration of attributes from some other visual cortex models, ICM utilized the attribute of latency in information transfer in biologic neuron system and the attribute of non-linear coupling modulation. ICM boasts the attribute of processing image without training a large amount of learning sample. ICM have the attributes of latency in information transfer in biologic neuron system and non-linear coupling modulation. These attributes have the processing speed and result advantages over the traditional image processing method in image noise suppression, image morphologic operation, and image segmentation. Therefore, ICM shows great research and application value in the scope of image processing.ICM contains two attributes, namely, the attribute of latency in information transfer in biologic neuron system and the attribute of non-linear coupling modulation. Directly inheriting the anatomic research result of the mammal's visual cortex system, ICM is closer to the actual biologic neural network and is more suitable for the image processing work. Meanwhile, ICM simulates the phenomena of synchronous spike 35 to 75 Hz oscillatory impulse stream when field of vision in mammal's visual neural system receives relevant stimulus. Moreover, ICM has the ability of compressing high dimension data into one dimension time pulse sequence, which is similar to the actual biologic vision neural network. However, ICM itself will produce the effect of auto-wave, the phenomenon of which is that the border of the object inside the image will emit the fake border in a diffused way accompanying the iteration in the ICM. The diffusion of fake object border will cause the severe interference to the post-object segmentation and recognition. Before our research work, the data that ICM can process is still 2D data; ICM has not the ability to process high dimension data.We analyzed the cause of autowave problem in ICM and proposed solutions to autowave problem. Our research was aimed at avoiding the pre-training process in the traditional neural network in processing the image, and we proposed some key image processing technologies based on the ICM. We proposed different ICM models for image noise suppression, image morphologic operation, and image segmentation. We implemented these proposed models and compared our models with national and international outstanding methods in different image processing tasks by experiments. The experiment results showed that our models were more efficient and accurate than these methods. We analyzed the cause of impulse noise in the image and proposed noise suppression mechanism based on ICM. The mechanism showed excellent result in noise suppression in the image contaminated by impulse noise and high efficiency in de-noise processing.We introduced ICM into the scope of image morphology, which demonstrated high application value in post image information measurement and pattern recognition.We solved two problems of inefficacy segmentation of spine in the X-ray gray spine image: one problem is that the image itself contains high background noise information; another problem is that the threshold of neurons in ICM must be manuallu initialized. We solved these problems by automated determination of the initial threshold value of the neurons in ICM on the basis of the maximization principle of the image segments' information entropy.We evolved ICM into 3D-ICM, which can handle high dimension data. We applied 3D-ICM into nature color image automatic segmentation, which is also based on the maximization principle of the image segments' information entropy.