Interactive Extraction of High-Resolution Remote Sensing Image Surface Feature Based on Full-Connected Conditional Random Field

An interactive extraction method of high-resolution remote sensing image surface features based on full-connected conditional random field is presented. Through human interaction tag estimation foreground model, combined with color and texture features, and using the simple linear iterative clustering(SLIC) algorithm to over segment the input image, the maximal similarity based on region merging (MSRM) is used to expand the foreground region and establish the global information of the full-connected conditional random field to describe image. The model inference is realized by the high-dimensional Gauss filtering method which is based on the mean-field estimation. The contours of the area features are obtained. The method is proved to be effective by experimental extraction of surface features such as water, woodland and terraced fields in high-resolution remote sensing images.