Visual computation multivariate connection model-based saliency contour perception method

Abstract

The invention discloses a saliency contour perception method based on a visual calculation multivariate connection model. The visual calculation model with a multivariate connection characteristic is constructed. On the basis of LGN feedforward connection, LGN neuron sparse coding characteristics are simulated, a weight factor is added, preliminary texture suppression is achieved, and a primary sensing result of a contour is obtained; on the horizontal connection of the primary visual cortex, simulating a primary visual cortex windmill-like structure receptive field, and adjusting the discharge intensity of a central neuron based on the distance between neurons and the optimal orientation included angle; and on the feedback connection of the advanced visual cortex layer, simulating the hue perception characteristics of the advanced visual cortex layer, constructing a three-channel hue perception model containing surround suppression, and obtaining the response of the advanced visual cortex layer to an image target or structure. According to the invention, through the construction of the visual calculation model with the multi-element connection characteristic, the obtained contour can effectively highlight the main body target while suppressing the texture noise.

Description

technical field [0001] The invention belongs to the field of visual neural computing, in particular to a saliency contour perception method based on a multi-connection model of visual computing. Background technique [0002] Contour perception is a key early step in image analysis, and its goal is to extract object contours as completely as possible while removing background and texture noise. The traditional contour perception method is mainly based on the difference of pixel color or gray value feature in the local neighborhood, and usually ignores the important role of the overall information of the target in the local contour detection process. More texture noise. [0003] Current contour perception methods that incorporate biological vision mechanisms, such as visual computing models that incorporate mechanisms such as direction selectivity and receptive field synergy, can effectively improve the interpretability and stability of contour perception models. However, th...

AI Technical Summary

Problems solved by technology

The traditional contour perception method is mainly based on the feature difference of pixel color or gray value in the local neighborhood, usually ignoring the important role of the overall information of the target in the local contour detection process, so the obtained contour perception results are easy to lose part of the structural characteristics of the target, and will introduce more texture noise

Images