Multiple cues region growing segmentation on tongue image

This paper presents an improved region growing algorithm for tongue image segmentation by integrating with symmetry, colour and texture cues. The symmetry cue is calculated by symmetry affinity matrix and considered as symmetry constraint. The colour constraint is computed by nonlinear transformation of HSV space, and the texture constraint is obtained by calculating a 12-dimensional texture features at each pixel based on Gabor filter with six orientations and eight scales. After combining these constraints as the pixel aggregation criterion for region growing algorithm, we propose a multiple cues region growing algorithm. Then, the segmentation experiments on the TCM tongue images proceed by selecting three seed points interactively. Meanwhile, the symmetry axis would be determined with these three points and used to symmetry cue calculation. Finally, compared with the previous C2G2FSnake algorithm and the single cue-based region growing on both qualitative and quantitative perspectives, the proposed algorithm shows the best segmentation accuracy and powerful stability, especially the tongue images with cracks and coatings.