Building extraction from panchromatic high-resolution remotely sensed imagery based on potential histogram and neighborhood Total variation

In order to extract buildings using only gray information, this article proposed an approach for recognizing and extracting buildings from panchromatic high-resolution remotely sensed imagery based on shadows and segmentation. First, shadows were detected by potential histogram function. Second, the value of neighborhood total variation for each pixel was calculated, and then binarization and annotation were implemented to generate lable regions whose centroids were used as the seeds of the region growing segmentation, candidate buildings were selected from the segmentation result with the constraint of aspect ratio and rectangularity. At last, shadows were processed with open, dilate and corrode operations respectively, buildings were extracted by computing the adjacency relationship of the processed shadows and candidate buildings, and the building boundaries were fitted with the minimum enclosing rectangle. For verifying the validity of the proposed method, eighteen representative sub-images were chosen from PLEIADES images covering Shenzhen, China. Experimental results show that the average precision and recall of the proposed method are 97.95 % and 79.40 % for the object-based evaluation, and are 98.75 % and 83.16 % for the area-based evaluation respectively, and it has more 10 % and 6 % increase in the overall performance for above two evaluation criterion comparing with two other similar methods.