An Improved Superpixel-based Fuzzy C-Means Method for Complex Picture Segmentation Tasks

Fuzzy c-means(FCM) has attracted wide attentions on picture segmentation as its fuzzy attribute matches the histogram distribution of a picture. However, the fuzzy c-means for the segmentation of a picture with massy noises is barely investigated. In this paper, an improved superpixel-based fuzzy c-means is proposed to segment a massy noise corrupted picture into more than two classes. Firstly, bilateral filtering is used to reduce the compact of noises and makes the picture smoother. Secondly an adaptive method is proposed to fuse the features of the original picture with filtered features. Thirdly simple linearly iterative clustering(SLIC) is used to detect the edge of the picture to avoid over-segmentation. Finally, the histogram-based fuzzy c-means is used to get the segmentation result. In the experiments, the results show the proposed method achieves a $0.004 \sim 0.014$ higher mPA and $0.004 \sim 0.06$ higher mIoU than other seven algorithms. Besides the segmentation results also show that the over-segmentation is reduced.