Abstract: A fast image representation method based on wedgelets is proposed in order to solve the problem that the image representation method by wedgelets has high computational complexity and storage space. The proposed method uses a recursive top-down quad-tree pruning algorithm compared to traditional bottom-up pruning strategy, based on fast multi-tree search and decorating leaves by wedgelets to achieve fast operation and reduce the storage space. Meanwhile, simple coding strategy and some search tips to improve the computational efficiency are presented and analyzed in this paper. Complexity analysis and Experimental results show that the method reduces computation complexity for searching different dyadic squares with desirable rate-distortion behaviour and captures natural geometric structure of image effectively.

Key words: wedgelets, multi-tree dictionaries, quad-tree pruning, coding, rate-distortion

摘要： 针对用wedgelets表示图像存在计算冗余和存储空间大的问题，提出一种快速的基于wedgelets的图像表示方法。采用与传统的自下而上的剪枝策略不同的四叉树剪枝算法，通过基于快速多叉数树搜索及仅用wedgelets表示树叶来实现快速运算和减少存储空间，并且提出了一些提高计算效率的搜索和编码技巧。复杂度分析及实验结果表明，该方法能降低计算复杂度且有理想的率失真性能，并有效地捕获图像的几何结构。

关键词: wedgelets, 多叉树字典, 四叉树剪枝, 编码, 率失真