| The visual functions of cortical neural network and its mechanismfor contour integration are firstly arranged, then quantitative analysis andsynthesis on spatial-temporal characteristics of the neurons in the corticalneural network are implemented, and finally a cortical neural networkmodel based on learning mechanism for contour integration is established,and a set of experimental results as well as its analysis are given.Considering that the biological visual mechanism of contourintegration is very complex, and as a subsystem in biological visualpathway it is not an independent component, the biological visualfunctions of cortical neural network and its mechanism for contourintegration are firstly arranged. On balancing the realization complexityand duplication fidelity of biological mechanism for contour integration,the topology of cortex neural network for target contour integration arethen determined. Based on quantitative analysis and synthesis onspatial-temporal characteristics of the neurons in the cortical neuralnetwork, neural model for contour integration are then established.Cortical learning as well as information selection mechanism arealso studied in this paper. Here the objective function for training thenetwork is constructed by curve smoothness and closure, and it is carried out by corresponding lateral neural network, by which traditional lateralinhibition mechanism was implemented.Finally, qualitative as well as quantitative analysis on experimentalresults and a set of comparisons with other method are given. Theexperimental results demonstrate that the proposed neural network modelfor contour integration can be used to grouping the discontinuoussegments in contour of an object, and also that learning mechanismimproves the neural network efficiency from the point view of contourintegration. Experimental results also confirm that the performance of theproposed model on contour integration is superior to the Canny operator. |
PDF Full Text Request