Lattice Vector Quantization for Image Coding Using Expansion of Codebook

Abstract

We present a method for image coding which uses expansion of basic codebook, based on energy of the vectors, for assigning lattice points to the incoming vectors. Two expansions of the basic codebook are used. Since in this method range or radius of the codebooks is based on the energy of the vectors of input image, for different test images we get different basic and expansion codebooks. If the radius of the codebook is very small then mapping of many points may be beyond the range of the codebook, if the radius is very large then time taken to find out the nearest lattice points may be more. So by using this method problem of lattice codebook overload can be bypassed as well as search of vector is reduced because instead of using one small or large codebook, codebook is expanded as and when required. Perceptual performance of peppers, barbara and goldhill images are shown with their MSE and PSNR. Execution time required getting reconstructed image using single codebook and expanded codebook is compared.