Local Shadow Compensation Considering Feature Differences for UAV Images

Shadows become evident in UAV images, which will cause contrast reducing, loss of details and texture blurring, resulting in the decrease of image quality and affecting the accurate identification of ground object information. The most advanced existing methods can correct shadow regions, but the color distortion is still evident and especially noticeable in areas with complex surface features. Therefore, we propose a novel method of local shadow compensation considering feature differences (LSCFD) for UAV images, which resolves the overlooked spectral heterogeneity inside shadow regions through feature-aware segmentation and statistical analysis. Firstly, the Mean shift algorithm is used to divide the study area into some small regions, and the isolated segmentation regions in the shadow regions are removed. Secondly, the non-shadow regions with the same boundary as the shadow regions are selected as the reference regions. Finally, the mean and standard deviation of the R, G, and B components of the shadow and reference regions are calculated respectively, and the compensation formula is used to compensate for the shadows. Comparative analyses demonstrate the LSCFD’s superior performance against five established methods, such as Wallis filtering, Region matching, Lighting ratio, and so on. The results show that LSCFD achieves the lowest values in comprehensive quality metrics including the sum of squared brightness deviations and squared gradient deviations, SAM and CIEDE2000 across test datasets and exhibits favorable robustness and adaptability, which can effectively suppress color distortion and improve compensation quality. Moreover, the compensated regions exhibit greater consistency with non-shadow regions, which indicates enhanced visual and quantitative accuracy.