【摘要】 为提高图像稀疏分解的速度和重构质量,提出一种基于核心原子库的图像MP稀疏分解快速算法。该算法首先利用核心原子库替代图像MP稀疏分解过程中的过完备原子库,可有效提高原子库的形成速度,且为最佳原子的全局搜索提供了可能;然后将图像和核心原子库中原子转换为一维信号,利用一维FHT寻找与待分解图像匹配的最佳原子,从而提高图像与原子的匹配速度。该算法不仅能有效提高图像MP稀疏分解的速度,而且克服了遗传算法局部最优的局限性,实现原子匹配的全局最优,相同条件下其分解结果具有更好的重构质量。实验结果验证了算法的有效性。更多还原

【Abstract】 In order to improve the speed and reconstruction quality of image sparse decomposition,the fast algorithm for image MP(matching pursuit) sparse decomposition based on FHT(Fast Hartley Transform,FHT) and a core dictionary is proposed.Firstly,this algorithm replaces the complete dictionary with the core dictionary.Such operation improves the speed of dictionary forming and provides the possibility for the global search of the best atoms.Secondly the given image is converted to the one-dimensional signal and the atoms in the core dictionary which is the compact form of the common dictionary are also correspondingly turned into one-dimensional.Then cross-correlation between the image or the residual images and atoms is calculated by one-dimensional FHT to find the best atoms.In conclusion,this algorithm can improve the speed of finding the best atoms effectively.Compared with the common methods of image sparse decomposition,which are based on MP or the genetic algorithm,this algorithm improves the decomposition speed and reconstruction quality efficiently under the same conditions.The results show the effectiveness of the proposed algorithm.更多还原