A Cooperative Relaxation-Based Method for Range Image Segmentation
Imene Belloum, Mourad Bouzenada, Smaine Mazouzi · International Journal of Computing and Digital Systems · 2024
In range images, edge detection is a hard task because of high levels of noise and distortions, produced at image acquisition.However, detecting edges in range image allows to enhance region-based segmentation, which is a key step for any further step in image analysis and understanding.To deal with the hard problem of range image segmentation, we propose in this paper a combination of an edge-based and a region-based methods for the segmentation of noisy and distorted real range images.The proposed combination is based on the relaxation of the detection results of both edges and regions.For region-based segmentation we have proposed a new multi-seed region growing algorithm using curvature as homogeneity criterion.For edge-based segmentation, we have detector adapted the Canny filter by taking into account a vectorial representation of raw data in range images.So, the two resulted sets of edges are obtained from two different representations of the same image data.The principle of the introduced method consists in matching the two segmentation results as a mutual self-regularization, and whose objective is to produce an optimal final segmentation with respect to a given criterion of optimality, expressed by a well appropriate energy-based objective function.The optimal solution, considered as unique, is calculated by the simulated annealing algorithm, and consists of the best edge map that can obtained on the processed range images.The experimental results, using the ABW database, show better results compared to those obtained by the classical Canny detector.So, we can conclude that the proposed relaxation-based combination allows more efficient segmentation of range images.