Summarization of paper "Compressed Random-Access Trees for Spatially Coherent Data"

最新推荐文章于 2022-07-04 19:35:48 发布

原创最新推荐文章于 2022-07-04 19:35:48 发布 · 457 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#pointers #compression #tree #encoding

论文总结专栏收录该内容

1 篇文章

订阅专栏

本文提出一种随机存取树的自适应多分辨率层次结构，该结构特别适用于表示空间连贯的图形数据，并通过减少存储空间和带宽来提高效率。方法采用原始细分树结构、节点指针替换为局部偏移量、森林状mipmap结构及层级残差的有损压缩等技术。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

1. Goals

Randomly-accessible tree is proposed for the adaptive multiresolution hierarchies which are highly efficient at representing spatially coherent graphics data. To reduce the space and bandwidth are the main goals of this method.

2. Method

In this paper, there some novel elements as follows:

l Use of a primal-subdivision tree structure.

l Replacement of node pointers by local offsets.

l A forested mipmap structure

l Lossy compression of inter-level residuals.

The main steps for encoding:

1) construct the mipmap in fine-to-coarse order and compute the desired values d_l with multilinear interpolation.

Constructing mipmap

Figure 1: constructing mipmap

2) compressing the tree data.

First, we need to compute the predicted value p_l from parents. Then, we have to compute the residual r_l = d_l - p_l which will be compressed using brood-base VQ resulting in compressed residual r_l ^c . At last, we should compute the codebook for the VQ tree.

Computing the predicted value

Figure 2: computing the predicted value

3) compressing tree topology.

We process each level of the VQ tree in fine-to-coarser order, looking to prune it’s leaves under the condition that accumulated approximation errors at all affected nodes do not exceed the tolerance т .

Pruning leaves

Figure 3: pruning leaves.

Moreover, we need to pack the tree topology by replacing pointers to local offsets for reducing spaces.

3. Conclusion

This method is not intended for detail color images, which result in dense trees and are thus better handled by block-based schemes.

This method starts with a complete tree and prunes leaves during compression. So the current tree construction procedure can only deal with small datasets.

Because this method replaces the pointers with local offsets, so it need more time for this process. Moreover this compression method is lossy.