A Dictionary-Classification based Super-resolution Reconstruction
This paper presents a new approach to single-image super-resolution,based on spares signal representation.Sparse representation modeling of data assumes an ability to describe signals as linear combinations of a few atoms from a prespecified dictionary.Inspired by super-resolution reconstruction via spares representation,we proposed a dictionary-classification based super-resolution algorithm.The main idea of our algorithm is to classify the training images in the process of dictionary training.In order to group the images in training library,we applied four masks to extract corresponding features of each image,then trained the dictionaries and sparse coding with the grouped images respectively.Before the process of reconstruction,we divided the image patches into flat area and non-flat area with a threshold of variance.Then we processed the patches in the non-flat area using the same method in dictionary training.Next,we searched the corresponding dictionaries in library according to the extracted feature to rebuild the patches in non-flat area.At last,we reconstructed the flat area with bicubic interpolation.We apply the dictionaryclassification based super-resolution algorithm to natural images and demonstrate that the recovered high-resolution image is competitive to images produced by other super-resolution methods but with faster processing speed.
dictionary-classification sparse-representation super-resolution reconstruction
Jingyuan Xie Feng Liu Zongliang Gan
Image Processing and Image Communication Jiangsu Key Laboratory, NJUPT Nanjing,China Image Processing and Image Communication Jiangsu Key Laboratory, NJUPT, Nanjing, China
国际会议
2012 IEEE 14th International Conference on Communication Technology(2012年第十四届通信技术国际会议(ICCT 2012))
成都
英文
1055-1059
2012-11-09（万方平台首次上网日期，不代表论文的发表时间）