Image Threshold Segmentation Based on Multi-scale Space Analysis

This paper proposes a method for image threshold segmentation using multi-scale space analysis.Multi-scale space is obtained through multi-scale filtering to original histogram of luminance.On the analysis of multi-scale space,the range of filtering scales is determined so as to avoid scales continually increasing/decreasing and improve the efficiency of algorithm.Then the original histogram is filtered by coarse-fine scale in the determined range of scales.Locations of significant inflexions at different scales can be obtained as thresholds by computing the second difference of the filtered histogram.Accroding to the number of inflexions at the largest scale and inflexions corresponding to graylevels,the fine locations of inflexions at the largest scale are obtained as thresholds.The result shows that the proposed algorithm is effective by contrast with other image segmentation algorithms.