Visual pathway multi-receptive-field directional association-based profile detection method

Abstract

The invention discloses a visual pathway multi-receptive-field directional association-based profile detection method. According to the method, classic receptive field characteristics of retinal neurons and non classic receptive field characteristics of LGN (Lateral Geniculate Nucleus) cells in a visual pathway are simulated from spatial scale characteristics of receptive fields of all levels and mutual associativity of receptive fields of the same level in the visual pathway, and rectangular receptive fields of the LGN cells are associated directionally for simulating direction selection characteristics of simple cells of V1 (primary visual cortex). According to the visual pathway multi-receptive-field directional association-based profile detection method proposed by the invention, an important role of visual characteristics in profile detection is brought into full play.

Description

technical field [0001] The invention belongs to the field of visual neural computing, and mainly relates to a contour detection method based on orientation correlation of multiple receptive fields of visual pathways. Background technique [0002] Contour is a key attribute of the main object in the image, and its accurate detection is of great significance for improving advanced visual performance such as target recognition and image understanding in the later stage. Usually contour detection needs to accurately locate the target, highlight the continuity of the subject contour, and suppress the background texture edge at the same time. However, in actual scenes, due to the existence of interference factors such as illumination, ghosting, and similarity of texture contours, traditional contour detection methods are difficult to meet the above requirements at the same time. Considering the outstanding recognition and comprehension ability of the visual nervous system, the cu...

AI Technical Summary

Problems solved by technology

[0003] The current contour detection method that simulates the visual mechanism generally starts from the perspective of a purely computational model, directly using image pixels as the operating object of the model, while ignoring the most basic electrophysiological activities in the visual system, that is, hundreds of billions of neurons. The intrinsic correlation between activity and visual cognition; (2) The current modeling of visual mechanisms is usually limited to a single level of the visual pathway, and most of them use a single neuron receptive field model to simulate the function of the visual cortex, even if multiple levels are considered The characteristics of the receptive field, usually only give a single "input-output" relationship, without taking into account the hierarchical correlation characteristics of visual information in the processing and processing of visual pathways and the physiological structural basis on which this characteristic depends, which is easy Causes loss or redundancy of visual information, which differs from subjective visual perception results

Images