Abstract: Aiming at the shadow detection problem of single image in complex environment, we proposed a fast automatic shadow detection algorithm based on multi-scale super-pixel fusion. Firstly, the depth image was used to calculate the normal vector and spatial coordinates of every point, and the simple linear iterative clustering algorithm was used to complete multi-scale super-pixels segment for the color image. Secondly, the shadow confidence algorithm was used to estimate the shadow confidence of the super-pixels in each scale combined with the chromaticity, the normal and the spatial position information of the image. Finally, the trained Adaboost classifier was used to fuse the super-pixels shadow confidence in each scale, and the final judgment result was obtained. The experimental results show that the accuracy of the proposed algorithm is significantly higher than the original shadow confidence algorithm, and the running time is about 10% of the original shadow confidence algorithm. It performs more prominently in the detection of small shadows, large shadows, and soft shadows with unclear edges, which is suitable for pre-processing of the images in complex light environment.

Key words: Adaboost algorithm, shadow detection, super-pixel segmentation, depth image