An improved initial codebook generating method-orthogonal increment splitting algorithm

This paper present a comparison of three main algorithms generating initial codebooks with reference to the convergence accuracy and speed. With both the elements considered, splitting methods are found to be more advantageous. Through a study of the common splitting method, an improved one-the orthogonal increment splitting algorithm is proposed. The new method can better utilize the symmetrical property of the speech or image sources, and endow the codewords of the initial codebook with a better distribution in the source space compared with the previous splitting method. The experimental results have confirmed the advantages of the new method, which can effectively eliminate the empty cells and non-typical codewords that may appear in the iteration process, thus achieving a better performance in both convergence speed and accuracy