The first 3-dimensional (3-D) coding scheme was presented in 1977. After a short prosperity, 3-D coding became ignored due to the popularity of the hybrid scheme. Recently, however, 3-D coding has been paid attention to because of its natural approaches. 3-D approaches using various coding schemes such as subband, fractal, wavelet and vector quantization(VQ) have been inquired for video compression. Among these, investigation of 3-D VQ has been tardy mainly due to the difficulty of making VQ codebooks for large vector dimension. In a previous work, 3-D cube of a fixed size and shape (spatially 4x4 block, temporally 4 frames) was used. In the fixed cube, however, it is hard to make efficient use of local activities of images at reasonable bit rate.
This paper proposes a new 3-D VQ method by using adaptive segmentation according to three temporal activity modes and an efficient codebook generation algorithm for large dimensional vectors. First, changed and unchanged regions are classified for blocks of large size by frame difference. The unchanged blocks in the current frames are just replaced with the contents of the corresponding blocks in the previous coded frame. On the other hand, changed or temporally active regions are consiered as 3-D blocks and categorized with respect to their temporal variances. First, adaptive temporal segmentation is applied based on temporal correlation of image samples, and then each subblock is spatially segmented by using spatial variances. With these adaptive partitions, local activities are reflected to the coding scheme in a more efficient and decent manner.
The performance of the proposed scheme is investigated by computer simulation. The proposed scheme is proven to provide improved image quality in comparison with the conventional method and to compete with the standard coding scheme(example; H.261) in performance.
