A new general-purpose color image segmentation editor for the purpose of extracting semantic object is designed, implemented and tested on a number of various natural scene images. Our system integrates deformable models and image statistics including intensity, color, gradient and texture. First, a low-complexity edge-linking algorithm is proposed as a post-processor to Canny operator, which basically builds regions of smooth intensity surfaces by closing all dangling edges. The topologically-based region labeling makes full use of relationship among pixels and produces good label image. To refine the extracted object of interest, the deformable model based on energy minimization is applied. Experimental results suggest the efficiency and accuracy of the system in segmentation operation.
展开▼
