A High-Resolution Remote Sensing Images Segmentation Algorithm Based on PCA and Fuzzy C-Means

Abstract

Aiming at optimal segmentation scale for different surface features with different features in high-resolution remote sensing images takes a lot of experiments and exists subjectivity. This paper proposes an optimal segmentation algorithm, a method that combines principal component analysis (PCA) with fuzzy c-means (FCM). In this method, the initial clustering centers of FCM are generated by sorting values after dimension reduction by PCA on high-resolution remote sensing images. Then using fuzzy c-means algorithm merges the homogenous image units into one object, and thus, we can gain the segmentation results which rule out influence of subjectivity and uncertainty of initial clustering centers and segmentation scale. Our final result, visual evaluation and clustering internal evaluation indicators and segmentation evaluation indicators show that the high-resolution remote sensing images segmentation algorithm based on PCA and FCM is better than original FCM, and other traditional image segmentation methods mentioned in the paper.