Shadow Detection on High-Resolution Digital Orthophoto Map Using Semantic Matching

Abstract

Shadow detection and compensation on high-resolution orthophoto is one of the most important tasks for ensuring high-quality of the radiometric balance of digital orthophoto map (DOM). This paper proposes a novel shadow detection method through semantic matching between the artificial “shadow” polygons (ASPs) and the real shadowed polygons (RSPs). The ASPs are created by digital building model (DBM), solar zenith and solar azimuth. A group of polygons semantic features, such as position similarity, area similarity, direction similarity and shape similarity are described and used as matching parameters, and then the real shadow regions (polygons) (RSPs) are detected using two-level matchings between the ASPs and the RSPs. The initial of matching aims at obtaining the initial probability and the candidates of the matching pair through determining the initial search circle cantered at the ASPs. The final of matching aims at finding the final match pairs through iteration of semantic matching, of which the maximum probability, which is calculated by the support coefficient of the adjacent match pair, is adopted as the criterion of the iteration. The experimental area located in Denver, Colorado, USA is used to validate the proposed algorithm. When compared with the dual-threshold, the Least-Squares Support Vector Machine (LS-SVM), and the UNet shadow detection method, the method proposed in this paper is able to increase the rate of shadow detection by 38.98%, 17.33%, and 13.14%, and decrease the fault detection rate by 19.43%, 10.01%, and 4.57%.