Contour and Boundary Detection via Perceptual Mechanisms of Primary Visual Cortex
|School||Huazhong University of Science and Technology|
|Course||Pattern Recognition and Intelligent Systems|
|Keywords||Contour and boundary detection Context interaction Environmental suppression Spatial enhancement Non-classical receptive field Primary visual cortex Visual perception mechanism|
Contours and boundaries define the outer shape of the target to determine the dividing line between the regions, they are humans and computers important feature of target recognition. However, from the chaos of the natural contours of the object in the scene to extract the boundary is a very difficult task. To do this, three main issues need to be addressed: 1. Exclude a large background texture generated by the local edge ingredients; 2 scene context information based on local ingredients will be organized into meaningful global features; 3 Some important structural lack of a clear physical definition (for example, the boundary of the texture), and some part of the target and background may be of the same strength, so that the boundary or the lack of response is very weak partial valid evidence. In response to these difficulties, the paper according to the primary visual cortex perception mechanisms established various contours and boundary detection model, and by combining images and natural images verify performance of the algorithm. In order to reduce the background texture edge ingredients and highlight the border region, this cycle using non-classical receptive field inhibition of the dynamic properties of texture presents a method for inhibiting. This method textures and borders to take a different way, which significantly reduces the background interference components meaningless, and selectively preserved isolated silhouettes and regional boundaries. How complex scenes with the same spatial structure will be organized into a significant component of the outline is another key aspect of the study. In this paper, concyclic rules and visual preference for low-curvature path defines a silhouette combined with local aggregation function that will work with the arrangement and orientation of the two properties with clever link. The interaction through the context of local ingredients will be integrated into a meaningful global features and highlights from the background. By spatially separated excitement and inhibition zone, this paper will enhance the environment and spatial suppression combined in a unified model, allowing two opposing perceived behavioral exist. Based on this model, this paper emphasizes the spatial enhancement and environmental boundary detection and suppression in silhouette play different roles, mainly in the surface and the inhibition of texture segmentation, which is mainly used to enhance the binding contours and graphics background isolated. Color images carry more than the gray-scale image information of the image and can help to produce better results. Therefore, to further expand the gray model to color image processing. Color model will involve more properties homogeneity suppression, can more effectively remove the texture edge; On the other hand, the color contours provide more information gathered, more conducive to the integration of the same attributes. Finally, two application projects - angiography image enhancement and synthetic aperture radar images of the road detection, indicating extensive use of the model.