Comparison of mean shift algorithms and evaluation of their stability

Abstract

Mean Shift is an universal and robust segmentation algorithm used also for image segmentation. Implementation corresponding to definition is slow for real-time image processing, thus some faster variants were proposed. Speed and quality of the segmentation can be influenced by the variant of the algorithm as well as by setting its parameters. Modification of a parameter causes change of number of segments and their shape as well. By using segmentation evaluation algorithm, we analyze and present these changes for original and modified Mean Shift algorithms. Influence of noise is also studied.