Image Segmentation using Multi-Coloured Active Illumination

Abstract

In this paper, the use of active illumination is extended to image segmentation, specifically in the case of overlapping particles. This work is based on Multi-Flash Imaging (MFI), originally developed by Mitsubishi Electric Labs, to detect depth discontinuities. Illuminations of different wavelengths are projected from multiple positions, providing additional information about a scene compared to conventional segmentation techniques. Shadows are used to identify true object edges. The identification of nonoccluded particles is made possible by exploiting the fact that shadows are cast on underlying particles. Implementation issues such as selecting the appropriate colour model and number of illuminations are discussed. Image segmentation using the proposed method, canny edge detection and watershed transform are performed on overlapping beach pebbles and granite. Evaluation results confirm that the proposed method is a successful extension of the original method and reveals its increased accuracy when compared with conventional segmentation techniques.