Aiming at image with high noise, a fast clustering segmentation approach combining peak testing is proposed in this paper. Two-dimension histogram was converted into one-dimension with parallel projection segmentation algorithm. Peak testing algorithm was used to test the clustering result of the pixels in noisy image. Parallel width in projection model was adjusted according to the testing result. As the histogram is approximated by double-peak distribution, noisy image can be segmented with weighted Fuzzy c-Means clustering algorithm. Experimental results proved that the proposed approach is fast and effective.