Research On Target Contour Detection Based On Fixation Eye Movement Mechanism

Contour detection is one of the basic and important tasks in machine vision,target recognition and other fields.Contour detection is different from the traditional edge detection,which suppresses the edge of the background texture and other meaningless,highlighting the edge of the target object,and is an optimization of the edge detection.Human or primate can quickly capture the target in natural scenes because of their visual system has the ability to quickly and effectively handle outside information.Bionic studies have found that the stimulate response of the Classical Receptive Field(CRF)can be suppressed by the non-classical Receptive Field(n CRF)in the primary visual cortex(V1)cells.The traditional contour detection algorithm utilizes the suppression properties of the receptive field,and proposes a circular texture suppression template which is simulated by two-dimensional Gaussian difference Do G function.With the visual perception mechanism gradually integrated into the field of contour detection,the attention of information processing mechanism in the visual system is increasing by research scholars.Studies have shown that even if we are staring at a fixed target,the eye is also still doing the small movement with an order of magnitude which the naked eye can not see.This movement can be referred to as fixation eye movement.The fixation eye movement of human has a self-repair and multi-directional micro-competitive characteristics,can correct the position offset which caused by the high frequency vibration and drift,the retina image to adjust to the optimal state.Inspired by this mechanism,mainly aim at the calculation aspect of the inhibition model,and based on the traditional Do G suppression template this paper proposes a new algorithm which generates eight sub-templates that based on the interval angle.And the new algorithm adopts the program which the value of corresponding angle area in the sub-template to replace the value of central area,to simulate the eye movement information processing mechanism.The final inhibition response is computed in competition with different response of sub-template.The inhibitory properties of the non-classical receptive field to the central neuronal light stimulation were simulated by the inhibition weight which contains the eye movement.At the same time,inspiration by the multi-level synchronization information processing mechanism in visual cortex,the multi-scale space theory is incorporated into the proposed eye movement inhibition template in this paper,and four scales of different receptive fields are used to detect the contour respectively.Then fusion the results from the four scales,and the final detection image is not only retained the contour information based on the small scale,but also make full use of the advantage large scale on the background texture suppression,and further improves the performance of contour detection and enhances the robustness of detection.In this paper,the new contour detection algorithm based on the fixation eye movement mechanism makes up the defect that traditional contour detection algorithms ignore the eye movement information processing mechanism,which is more consistent with the physiological characteristics of cell receptive field.Using the images in the natural scene to test experiment,which from the Ru G database and the Berkeley image segmentation database.The classical evaluation performance P and the Berkeley evaluation standard PR were used to validate the feasibility and validity of the algorithm.The experimental results show that the algorithm proposed in this paper considers the eye movement effect,obtains better contour detection performance,enhances the weak contour information,and suppresses the background texture to the maximum extent,which is a new idea in the target contour detection method.