Remote Sensing Image Segmentation Based on Mean Shift Algorithm with Adaptive Bandwidth

Abstract

Image segmentation is an important step in bridging the semantic gap between low level image interpretation and high level information extraction. Many image segmentation algorithms are available, i.e. active contour method, watersheds method, edge based method, threshold method, etc. Most of these algorithms are parametric and require the image with strong gradient. Mean shift algorithm is a non-parametric density estimation algorithm, which is popularly used in image segmentation recently. However, one bottleneck of the mean shift procedure is that the results of segmentation rely highly on selection of bandwidth. We present an improved mean shift algorithm with adaptive bandwidth for remote sensing images. The bandwidth of each pixel is adaptively adjusted according to the corresponding probability distribution. Compared with traditional fixed bandwidth, our proposed algorithm is both with high efficient and accurate in segmentation of high resolution remote sensing image.