摘要
提出一种基于学习字典的图像类推方法,较好地增强了图像类推的算法效率。先将样本图像对分块,统一进行稀疏编码,训练学习字典,以建立它们之间的稀疏关联,再将这种关联作为先验知识来指导图像类推。该方法主要有训练学习字典和类推重建两个过程。字典训练过程可离线实现,提高了计算速度,并且可实现大量样本的训练;在类推重建过程中,该方法将通用图像类推方法中的搜索、匹配过程转换为稀疏先验的线性优化问题,显著提高了算法的计算效率。通过纹理数值化、风格化滤波等图像类推实验,证明了方法是快速有效的。
To improve the computational efficiency of image analogies,this paper presented a novel image analogies method based on learned dictionary.The method first segmented sample image pairs to patches,which were unified for sparse coding and training learned dictionary.Then built the sparse association between the patch pairs,and defined as a priori knowledge for image analogies.The method mainly included two processes:training learned dictionary and image analogies.The dictionary training process could be off-line achieved to improve the computation speed,accordingly realized numerous samples training.During image analogies process,this method used the linear optimization problem of sparse prior instead of searching and matching in general methods,and improved the computational efficiency remarkably.Experiments with texture-by-numbers,stylized filter,etc.show the high efficiency of our method.
出处
《计算机应用研究》
CSCD
北大核心
2011年第8期3171-3173,3177,共4页
Application Research of Computers
基金
中国博士后基金资助项目(20080441198)
电子科技大学青年科技基金重点资助项目(JX0804)
关键词
图像类推
稀疏表示
学习字典
L1范数
image analogies
sparse representation
learned dictionary
l1-norm