Adaptive growing and merging algorithm for image segmentation
Hsuan-Yi Ko, Jian–Jiun Ding · 2016
Image Segmentation plays an important role in image processing as it is at the foundation of many high-level computer vision tasks, such as scene understanding and object recognition. In this paper, an adaptive growing and merging algorithm is proposed to segment an image accurately. First, mean shift is applied to produce superpixels, and then superpixels grow according to their lab histograms and textures under the constraint of the edge's intensity. In adaptive region merging, we use the proposed dissimilarity measures, which are based on colors, textures, region sizes and multi-scale contour maps with non-constant weights that are adaptive to the region features. Furthermore, we take account of the contact rate of two adjacent regions to avoid over-merging. We also exploit the saliency map to maintain the main objects when the number of regions is small. The simulations on Berkeley segmentation database show that our proposed method outperforms state-of-the-art methods.