Weak Image Edge Detection Technique And Its Research On Application Based On Visual Encoding Mechanism

Rich details are contained in the image edge which is significant to subsequent crucial tasks,such as feature extraction, object recognition and so forth. While the edge results will loseamounts of information if the images with abundant details are processed by the traditionalmethods, considering the excellent processed effect by the visual system for the image edge, wetry to detect the edge with the visual encoding mechanism. Recently, though lots of researchershave explored and experimented in the field of edge detection from the perspective of the visualencoding mechanism, many stimulate the visual mechanism in the black box withoutimplementing the inner visual encoding characteristics. Hence, we study the characteristics ofvisual receptive field, orientation selectivity and so on first, then construct the network of thelattice neuron populations with incorporating the visual encoding characteristics mentioned above,and the colony images with rich edge details are taken as the application object for explaining theeffect of visual encoding characteristics in edge detection; afterwards the STDP mechanism andthe color encoding mechanism are considered to play a role in the edge detection; last, inhibitorysynapse is adopted as the synapse’s mathematical model for improving the computingperformance, meanwhile we introduce the mechanism of selective attention to serve for the edgedetection, and we get better results than the traditional methods through the edge images beingweighed by the indexes of ROC and information entropy. The main tasks and research results arelisted as follows:(1) A method of edge detection in colony images based on spatial and temporal information ofresponses from the lattice neurons was proposed. We studied the effect imposed by visualencoding mechanism including visual receptive field, orientation selectivity, rank order and soforth on the edge detection, implementation steps were given, and Log-Gabor filter wasemployed to stimulate the characteristic of orientation selectivity in the visual cortex, finallywe got the results of edge detection through the constructed lattice neuron populations. Theresults showed that the image edge was detected effectively, whose strength could berepresented by the method proposed.(2) Considering the importance of synapses to the information processing in the network ofneuron populations, we studied the STDP mechanism of synapses, moreover, the color featureof the original image contains abundant edge details, a method of edge detection based on theSTDP mechanism of synapses in the visual cortex was proposed to detect the edge with the mechanism of color encoding. The results showed that edge acquired reveled abundant details,which were in accordance with the human’s perception laws.(3) Due to the inefficiency of the model for the STDP mechanism, and the situation that manycharacteristics which contains edge information hasn’t been made fully use of, a method ofedge detection based on spike coding of multilayer neuronal population with inhibitorysynapse was proposed. Inhibitory synapse was applied for interaction among neurons,meanwhile, the mechanism of selective attention was adopted to get focused areas as theprime objects of the edge detection, and then edge detection could be accomplished with theconstructured multilayer neuronal population. The results showed that we could get betteredge images than that of processed by the traditional methods, meanwhile, the integrity andthe abundance in the edge images were weighed better through the quantitative indexes ofROC and information entropy.