Road extraction in vague images on gray scale consistency and improved MSR and D-S evidence

Abstract Road detection on aerial and remote sensing vague images is a hard task. In this paper, an automatic road method for the proposed. The firstly uses improved MSR algorithm to enhance image, it automatically takes different scales in image regions, based depths obtained by dark channel prior algorithm. Then enhanced roughly segmented principle of local gray scale consistency, that, eight-neighborhood template considered as processing unit which threshold utilized all neighboring pixels detecting pixel. Finally, Dempster-Shafer (D-S) evidence features applied finalize tracing binary where, include length, width, aspect ratio fullness rate, parameters are least external rectangle segment, then detected roads regulated. experiments, 300 were selected testing, comparing several traditional algorithms recent semantic methods, testing results show that new satisfactory, accuracy up 89%.