• Most volume visualization methods assume, implicitly or explicitly, the main memory holds the whole volume data during run-time processing. But, it is not easy to use workstations or personal computers with several giga bytes of main memory yet.
• When we consider a method which repeates visualization of each subblock and composition of the images, It takes too much time to swap the subblocks in computing.
• Without a high performace computer, it is almost impossible to interactively visualize very large volume data.

• High Compression Ratio : The very large data must be compressed less than the main memory in size. In this case, we assume the size of memory ranges from 128 to 256 MBytes.
• Fast Decoding : Since voxels used in visualization are decoded from compressed structure, it is necessary to design effective encoding scheme to support fast decoding.
• Random Access : When we reconstruct a voxel with index (i, j, k), what other adjacent voxels have to be reconstructed is very inefficient. That is, we must be able to access only needed voxel randomly.
• Degradation Minimization : We generally use lossy compression to get high compression ratio, but must maintain good image quality as possible. In particular, the medical data used in diagnosis have to be reconstructed with almost the same quality as original data.

     Muraki compressed 3D volume data by wavelet transform in [11] and [12], but he made no mention of decoding time, random accessibility or the size of needed main memory in visualization process. And the quality of reconstructed images was not good. For fast decoding and random access, a method based on vector quantization was proposed in [13], but it didn't show high compression ratio nor fidelity in image quality. A technique based on Laplacian pyramid and vector quantization was presented in [4], but its results were also not good in image quality. A method used DCT(Discrete Cosine Transform) extended to 3D for volume data compression and visualization was introduced in [23]. It showed high compression ratio and fast decoding time. In this method, block-wise access was possible but pure random access was impossible.

     We have developed an effective 3D compression scheme for the CT, MRI and RGB data of the Visible Human that exploits the power of wavelet theory. In designing our compression methods, we have compromised between two important factors: high compression ratio and fast run-time random access. The first version of our encoding schemes ([9], [10]) has been improved in terms of both compression ratio and access time. The new scheme proposed in the paper provides much faster run-time data access as well as fairly high compression ratios~(up to 29:1 for CT, 112:1 for RGB) for the Visible Human dataset. Thus this scheme can be effectively used in interactive visualization system. In this paper, we introduce our 3D wavelet-based compression schemes, and describe applications we are currently building.

      The rest of paper is organized as follows: We introduce wavelet-based 3D compression in section 2, and describe the enhanced compression scheme in section 3. In section 4, we estimate experimental results and analyze. Next, some applications are introduced in section 5. Finally, section 6  includes conclusions and future works (Figure 1).