In computer vision, a feature is a locally detectable pattern of pixels from an image which may represent a piece of higher-level information about the image. Since most of the information in an image lies on the boundaries between different image regions, the edge (or curve) based features play an important role in computer vision. Edge feature extraction is always an initial step for numerous image understanding applications.; The conventional curve feature extraction methods rely heavily on the precise calculation of curve equation parameters or curvatures, which tend to be computationally intensive and less robust in terms of handling noises and curve shape distortions. This thesis developed a computational framework for curve feature extraction based on a perceptual organization model: Perceptual Curve Partitioning and Grouping (PCPG), which furnishes a more qualitative and symbolic way to detect and describe curve features. The framework contains two subsystems: curve partitioning, and perceptual feature classification. In the curve partitioning process, the system detects Curve Partition Points (CPPs) by applying the PCPG model, in that both local and global information of edge trace are considered, meanwhile, the partitioned segment geometry is used as well.; The proposed system has been tested on a set of images crossing the applications from man-made object recognition to medical vessel detection. The results demonstrated that the developed system can provide more precise result of perceptual edge features in a computationally efficient way.
展开▼
