Research On Image Feature Point Detection Technique Based On Biological Vision Mechanism

Feature point is the key information in the image which is significant to subsequent crucial tasks,such as image matching,image mosaic and so forth.The traditional methods are based on the change of gradient in brightness and color,while the feature point results will lose and misjudge amounts of information if the images with abundant details are processed by the traditional methods.Considering the excellent processed effect by the visual system for the significant information,we try to describe the visual information about feature point with the important visual biology mechanism.Hence,we study the dynamic characteristics of visual receptive field,feedback of visual cortical neurons and visual attention mechanism,then combine the sequence coding of neurons,the architectural images with amounts of feature points are taken as the experiment object,and we take the CCN as the evaluation index,the validity of visual mechanism in feature point detection was verified.The main results of research and work are listed as follows:?1?A method of feature point detection based on the self-adaption receptive field of the primary visual cortex was proposed.We researched the effect imposed by the visual mechanism,such as the dynamic characteristics of visual receptive field,sequence coding and visual attention mechanism,and the research steps were listed,the pixel information of image has a decisive effect on the feature points,so we constructed the adaptive receptive field according to the brightness information in different region of the image,and the feedback of different neurons in the primary visual cortex,then we got the feature points through the above steps,the problem of inaccurate feature point detection caused by the fixed model was solved.We came to a conclusion,the feature points were detected effectively and precisely,and the stability was farthest,taking the consistency of image feature point that named CCN as the evaluation index,then compared with the results of the traditional methods,the stability of this method was the highest,and the exponent level was between 10-1 and 10-7.?2?A method of feature point detection based on the self-adaption visual color information was proposed.Considering the importance of feature information contained by the three components of color in the image processing,we restructured the three components of color through the metabolic relation to detect the feature points.Considering the inhibition and stimulation of the peripheral neurons to the central neurons,then combined color restructuring with the effect of peripheral neurons and receptive field imaging to detect the feature points,the problem about missing the feature points caused by the change of color information was solved.The results showed that the feature points were detected precisely and it controlled the redundant points better,then compared with the traditional methods,the CCN of this method was the highest,and the exponent level was between10^-1and10^-6,so it was closer to the cognition of human visual system.?3?A method of feature point detection based on multiple orientations of edge was proposed.Considering the orientation of the texture and edge in the image,the problem about accumulation of feature points is serious in the edge,and the neuron spiking is various,we combined the orientation of the texture and edge with neuron spiking to detect the preliminary feature points and remove the redundant points,then utilized the self-adaption receptive field,neuron spiking and orientation of the edge to detect the representative features through the visual overlay mechanism,the problem about accumulation of feature points in the edge was solved.The evaluation index showed that this method ensured the real feature points and controlled the number of redundant points better,and the exponent level was between10^-1and10^-3.