ADAPTIVE SEGMENTATION AND BOUNDARY DETECTION FOR IMAGES WITH COMPLEX LIGHTING ENVIRONMENT

摘要

A new technique for the automatic extraction of objectrnregion and boundary from images with complex lightingrnenvironment is presented in this paper. Segmentationrnprocedure begins with the computation of an optimumrnthreshold to distinguish the darker regions in the image.rnThe image thresholding problem is formulated as anrnadaptive technique that progressively thresholds thernhistogram of an image. The thresholding algorithm usesrnthe spatial distribution characteristics of an image andrnautomates the thresholding process. The center of mass ofrnthis thresholded region acts as a seed for furtherrnprocessing. Then the exact object region is obtained byrnusing an Iris filter, which operates in the gradient space ofrnthe image. A quad-structure based technique is used tornenhance the speed of region search significantly. Arnboundary thinning and connecting algorithm based on thernapplication of a novel search window on the preliminaryrnboundary is also presented to obtain a connected singlernpixel width object boundary. The new method does notrnneed a priori knowledge about the image characteristics.rnThe performance of the new algorithm is evaluated byrnextracting the lumen region and boundary fromrngastrointestinal images. One of the main advantages ofrnthe new procedure is its high-speed response, whichrnmakes the real-time extraction of the lumen feasible.