Image segmentation remains as one of the most important tasks for image analysis and understanding. It deals with raw images in order to prepare them to be usable in automatic high-level processes, such as classification or information retrieval. We present in this paper a new adapted edge detector for range images. Its principle is inspired from the Canny detector so the inherent features of range images will be considered. Usually, Canny detector is used with greyscale or color images, where its direct application with depths does not provide satisfactory results. From the raw image, contained measured depths, a relief image that consists of an image of normal vectors to the local surfaces is computed. So, angles between neighboring vectors are used to compute an angle-based gradient. The latter is integrated in the Canny algorithm, so an edge map is produced for the range image. Real images from the ABW database were used in experimentation where the proposed new detector has outperformed the original Canny one by a ratio of 18 %.
Key words: Segmentation, Edge-Detection, Canny Detector, Range Images.
|
