Fuzzy split and merge for shadow detection

Abstract

Presence of shadow in an image often causes problems in computer vision applications such as object recognition and image segmentation. This paper proposes a method to detect the shadow from a single image using fuzzy split and merge approach. Split and merge is a classical algorithm used in image segmentation. Predicate function in the classical approach is replaced by a Fuzzy predicate in the proposed approach. The method follows a top down approach of recursively splitting an image into homogeneous quadtree blocks, followed by a bottom up approach by merging adjacent unique regions. The method has been compared with previous approaches and found to be better in performance in terms of accuracy.

Keywords

• Shadow detection
• Split and merge
• Fuzzy predicate
• ANFIS