ANALYSIS OF SKELETONIZATION ALGORITHMS’ FOR BYNARY IMAGES
Binary images skeletons constructing algorithms have been studied. The modern definition of a plane figure skeleton was clarified. The review and classification of existing skeletonization algorithms skeletonization has been performed. Algorithms were classified according to the technique of image processing inside the frame and according to the applied principles of skeletonization. There was marked that the skeleton of a plane figure constructing is Hadamard's ill-posed problem. Incorrectness of the problem produces the faults in resulting skeleton of two types - the formation of false skeleton branches and a violation of its continuity. Two quality indexes of the skeleton that cha-racterize these faults have been introduced for a quantitative comparison of different algorithms. The computing speed has been selected as another measure for comparing the performance of different algorithms. The computing speed is estimated with time of the skeleton constructing procedure for a given algorithm. This estimation should be carried out on the same hardware and software platform to provide comparability of different procedures. Five of the most popular algorithms were chosen to compare the quality of skeletonization: classical morphological skeletonization, two parallel iterative algorithms - Zhang-Suen's and Guo-Hall's ones, iterative sequential Stentiford's algorithm and high quality sequential algorithm by Schepin-Nepomnyaschiy. The algorithms have been implemented as a software in C ++ using OpenCV library in a cost-average performance computing platform for a comparative analysis. The frames of video stream of human moving with the size is 640 * 480 pixels obtained by the consumer webcam were used as original images for skeletonization. A comparative analysis was conducted from the view of the applicability of algorithms in the remote motor rehabilitation of patients. The figure skeletonization algorithm inside the frame is the most resource-intensive computing operation in such systems. Therefore, the purpose of the comparative analysis was skeletonization algorithm selection that works in near-real time and provides a sufficiently high quality skeleton for further patient's movement analysis. The analysis was conducted for processed 340 images from the motor rehabilitation database. It was established that a compromise for motor rehabilitation systems is the Zhang-Suen's algorithm. This algorithm combines high speed with high quality of produced skeleton.