Research And Emulational Implementation On Range Image Segmentation Algorithm

Range images are gaining popularity in different fields since they permit the efficient acquisition and representation of 3D information. They convey more knowledge about observed scene as compared to intensity images, and open new possibilities to image interpretation.Image segmentation is one of the main challenges in image analysis, and depends on correct feature extraction. Range image segmentation algorithms can be broadly classified into two categories: edge-based and region-based segmentation. Region-based approaches group pixels into connected regions based on homogeneity measures, while boundaries between regions are located by edge detection methods.Both techniques have their strengths and drawbacks. Edge detection is mostly criticized for its tendency to produce non-connected boundaries. Extensive post processing may be needed to provide the final segmentation. Despite of the guarantee of closed regions, region-based techniques suffer from a number of problems. Usually, they have complex control structures. In addition, commonly used region-based techniques such as region-growing and clustering have several critical design issues to be deal with. The performance of most region-growing approaches crucially depends on the selection of initial regions. In clustering-based methods it is difficult to adaptively determine the actual number of clusters in range images. Often, an over segmentation is achieved and a subsequent merge step is needed to provide the final segmentation.In this paper we propose an improved segmentation algorithm for range images, integration of edge-based and region based segmentation algorithms. Our work was partially motivated by the desire to overcome the above drawbacks inherent to most of the algorithms known from the literature. On the one hand, it could achieve better performance and connected boundaries compared with the classic range image segmentation algorithms; on the other hand, it gains much more time than region-based approaches.Our technique consists of three stages. First, an edge map is generated by a novel segmentation algorithm. The algorithm provides edge strength measures that have a straightforward geometric interpretation and supports a classification of edge points into several subtypes, such as jump edges, and crease edges. Careful examination of range images reveals that the usual definition of jump edges as discontinuities in depth values is not always adequate. Therefore we introduce a new definition of edges and present an original segmentation algorithm based on it. The second stage is to evaluate the obtained edge map according to the guide lines we've made in advance. If it comes well as a result, the achieved edge map is defined as the output image. Otherwise, the final stage is an innovatory approach to the classical contour extraction problem adopting morphological method. It shows a difference with the previous approaches which use the enclosed surface information. With the suggested technique, boundaries are obtained by using only the information contained in the edge map as to save much time.We have carried out extensive experiments using range images in a popular range image database. The results are compared to the four other traditional range image segmentation algorithms, demonstrating the efficiency of the proposed algorithm.