Study On Biological Visual Perception Calculation Model Oriented Image Processing Applications

The visual perception system plays an important role in the human cognitive function ofunderstanding the objective world. Modeling and applying for visual perception system from theperspective of neural computation, which has an important significance for the research ofinformation technology and artificial intelligence technology. Currently, the improvement ofneurological system research device allows people to take a furtive glance into the process ofvisual information flow detection, delivery, and coding. But researchers more study and analyzethe new features of visual perception system from neurophysiologic perspective. Our paperattempted to take advantage of some important characteristics of biological visual perception toachieve practical application of image processing. Firstly, the model based on cascade-bistableand array-cascade FitzHugh-Nagumo (FHN) neuron model were applied to utilize a certainintensity noise can enhance the biological visual characteristics of perception of weak signals, andexplain the mechanism of stochastic resonance plays an important role in the visual perceptionsystem function. Second, the model based on the orientation response Integrate-and-Fire (IF)neuron was proposed to achieve the application in image edge detection from the perspective ofthe characteristics of neurons orientation selection and the mechanism of impulse release andencode, and explain the key role of the characteristics of orientation sensitivity and impulserelease in the visual perception. Finally, low-dose lung CT images were used for practicalapplication of image enhancement with stochastic resonance mechanism and image edgedetection with neurons orientation sensitive character. The main work and research results of thisthesis are summarized as follows:(1) A new weak signal enhancement method based on stochastic resonance mechanism of cascadebistable model was presented. The thesis studied on the stochastic resonance signalenhancement ability of cascade bistable model on the one-dimensional analog signal. Themethod of selection the optimal parameter was given by using a quantitative evaluation index.The Raster scanning method was used for image dimension reduction to solve the inputdimension problem in image enhancement of row-column cascade bistable model. The ideaand implementation steps of image enhancement of cascade bistable model were given in thispaper. The experimental results showed that the propose method had a low-pass filtercharacteristic, removed the signal peak burr effectively and highlighted the contours of signal.(2) Consider the array characteristic of visual cortical neurons in the information flow transmission and processing based on the cascade characteristic, a new weak signalenhancement method based on stochastic resonance mechanism of array-cascade FHN neuronmodel was proposed. The thesis researched the stochastic resonance response of array-cascadeFHN neuron model on the one-dimensional signal with noise. The method combined withraster scanning and Hilbert scanning was used image dimension reduction to retain the spatialstructure correlation properties of image. The basic idea of application in image enhancementwas given in this paper. The propose method taken advantage of synergy effect betweenarray-cascade FHN model and noise. The experimental results showed that proposed methodcould make the output signal smoother, reinforce signal contour edges and details, reducenoise effectively, and improve the robustness and self-adaptation to internal noise.(3) For the characteristic of incentive orientation selection in the neurons response of visualperception, a new method of image edge detection was presented. To fully reflect orientationselection characteristic of human visual system, the Log-Gabor orientation response modelwas put forward to image multi-orientation decomposition. The characteristics of impulserelease and encode in the IF neurons model were used for low-dose lung CT image edgedetection. The experimental results showed that our propose method could enrich imagedetails, improve image edge characteristics, and enhance image edge layering by comparisonwith the traditional image edge detection methods.