Abstract:

An image color quantization algorithm base on peer group filtering (PGF) and vector quantization (VQ) is proposed. Firstly, PGF technology is used to filter image in LUV space, which smoothes image and maintains edges and details. And then, the local maximum of the 3D color histogram of the filtered image are chosen as the VQ codewords to make the image quantized. A quantization distortion function is defined, it takes visual characteristics into account. Split of cluster with the highest quantization distortion is iteratively performed until the requirement of total distortion is satisfied. Finally, agglomerative clustering can be applied to merge close clusters if further reduction of number of quantization colors is desired. The experiment shows that the objective and subjective quality of image produced by our algorithm were obviously better than that classical K-means algorithm.

Key words: LUV color space, color quantization, vector quantization, cluster