Abstract
Using color palette to solve the image segmentation problem can reduce the image color complexity and the difficulty of the computation efficiency, the main challenges in the work are selecting the representative color of the palette and how to refine the result. In our color palette method (CPB) we accept the color information from the blurred image which is different from contour-guide color palette (CCP). Using mean-shift (MS) clustering to find out the key regions’ color of the image and treat these color as representative color of the palette. We use the palette to repaint the whole image and followed by the segments refinement to improve the system’s result. The refinement includes two processes: 1. Segments redefinition which refers the image’s strong edge to split and redefine the region, 2. Region merging includes three steps “strong edge category”,” color category” and the “micro region merging”, In our refinement we not only consider the color space restriction but also add the contour clue. In this way we can easily solve the common problem in image segmentation. The performances of CPB are compared and analyzed with the CCP which also utilize color palette to solve the segmentation problems. We use Berkeley Segmentation Dataset for image segmentation test. Furthermore we use Microsoft Research Asia dataset to test our result in figure-ground segmentation.