An effective codebook initialization technique for LBG algorithm using subtractive clustering
Behzad Mirzaei, Hossein Nezamabadi–pour, Dariush Abbasi‐Moghadam · 2014
Vector quantization (VQ) is a lossy compression technique that mainly includes three stages: codebook generation, encoding and decoding. The efficiency of VQ extremely depends on the achieved codebook quality. The most commonly used method for VQ codebook generation, is the Linde-Buzo-Gray (LBG) algorithm. High sensitivity to initial codebook, is mentioned as one of the drawbacks of LBG algorithm. An effective codebook initialization technique in LBG algorithm has been proposed in this paper. The subtractive clustering has been employed to generate a proper initial codebook. Experimental results show that compared with other methods like common LBG algorithm and cluster density method, less RMSE and higher PSNR, is achieved owing to use presented method.