This paper presents a Component-Interlaced Vector Quantization (CIVQ) scheme for color image data compression. The CIVQ utilizes the inter-component correlations and local color features of image signals to reduce the computational complexity and to improve the coding performance. The three k-dimensional component vectors of a color image are subsampled and spatially interlaced into a single k-dimensional vector. And then, vector quantization is performed on this newly formed vector set. The receiver directly obtains the three quantized component vectors from the 3k-dimensional codebook which is designed by using a nonlinear interpolative technique. Two CIVQ experiments are conducted by using the mean-separated VQ and predictive VQ, respectively. Meanwhile, two color coordinate systems, RGB and YUV, are considered as well. The computer simulation results show that, in the case of k=16, a good quality of the reconstructed image can be achieved at 0.5 bit per color pixel.
抄録全体を表示