An Interactive Shadow Removing Tool: A Granular Computing Approach

This work proposes a tool to remove shadow from colour images with the help of user interaction. Shadow detection and removal is an interesting and a difficult image enhancement problem. In this work, a novel granule based approach for colour image enhancement is proposed. The proposed method constructs a shadow classifier using a Granular Reflex Fuzzy Min-Max Neural Network (GrRFMN). Classification and clustering techniques based on granular data are up-coming and finding importance in various fields including computer vision. GrRFMN capability to process granules of data is exploited here to tackle the problem of shadows. In this work, granule of data represents a group of pixels in the form of a hyperbox. During the training phase, GrRFMN learns shadow and non-shadow regions through an interaction with the user. A trained GrRFMN is then used to compute fuzzy memberships of image granules in the region of interest to shadow and non-shadow regions. A post processing of image based on the fuzzy memberships is then carried out to remove the shadow. As GrRFMN is trainable on-line in a single pass through data, the proposed method is fast enough to interact with the user.