Extended scheme of chan-vese models for colour image segmentation

The classic Chan-Vese (CV) model has been adopted in many applications. Many generalisations have been developed to improve both its applicability and efficiency, such as the two-phase model for vector-valued images by Chan and Vese. The vectorial CV model integrates multichannel information using the method similar to transforming a colour image into a grey one. It is invalid when an object and its background have close intensities. In this study, the classic CV model is generalised for colour images by using the strategy of segmenting an image from channel to channel. A multichannel segmentation combination (MSC) method is proposed to integrate the information of multiple level sets. In order to overcome the weakness that the correlation among different channels is not well considered in usual from-channel-to-channel methods, a novel multichannel ratio transformation (MRT) is introduced. And a variant HSV (VHSV) colour space is proposed to make every channel reflect region information without distortion. The experimental results show that the proposed scheme can obtain segmentation more accurately, and affords advantage in time-cost. Besides, the proposed method is valid only in the case of colour images with eight segments, but it can be enhanced by using the multiphase model.