Histogram Constraint Based Fast FCM Cluster Image Segmentation

Fussy c-mean cluster algorithm (FCM) is often used in image segmentation, but most FCM algorithm is time wasteful, for the purpose of improving segmentation efficiency, a fast segmentation algorithm based on histogram constraint is proposed. The new algorithm resample initial image to reduce data size, but reduction of data size space may cause distortion and make FCM converged to error threshold, in order to get best segmentation result, constraint based on distance deviation of histogram is incorporated. The initial histogram is smoothed to get its profile before calculating distance deviation, and its amplitude is transformed to a unified value. Correct resample ratio can be calculated from the changed histogram by golden section searching algorithm. Experiments are performed to validate the new fast FCM and the results shown that the segmentation result keeps in the same level with traditional FCM algorithm while the processing speed increases greatly. Compared with the other algorithms, the consumed time of our method is 1.0%-4.8% of traditional FCM algorithm, 3.4%-9.4% of 2D entropy algorithm and 5.1%-13.6% of Otsu algorithm, the average processing efficiency is 63, 22 and 15 times higher than above algorithms. The experiment results are consistent with theory and prove that distance deviation of histogram can reflect distortion degree of resample image, and the right segmentation can be deduced from reduced data.