Image super-resolution based on clustering and collaborative representation
The problem of image super-resolution (SR) reconstruction was first proposed by R.Y. Tsai and T.S. Huang in the 1980s. Subsequently, it gradually became a research hotspot in the field of computer vision and image processing, which attracted the attention of researchers. Image SR refers to the reconstruction of a high-resolution (HR) image from single or multiple observed degraded low-resolution (LR) image for the purpose of improving images visual effects and getting more available information. The applications of image SR reconstruction are extremely broad, which has been widely used in medical imaging, public safety, military and remote sensing systems. This technology can enable the image to transform from the detection level to the recognition level, or even further to the identification level.
SR reconstruction results of image “PPT”.
(a) Original; (b) Bicubic; (c) Zeyde; (d) NE+LLE; (e) NE+NNLS; (f) SF; (g) ANR; (h) Proposed
Professor Wang R G from the Institute of Computer Applications at the School of Computer and Information, Hefei University of Technology, proposes a novel super-resolution reconstruction method that combines image feature clustering and collaborative representation. Their method first clusters the features of the training samples, which gathers the image patches with the same features into a class, and then learns a complete dictionary for each subclass and obtains a mapping matrix for each subclass dictionary. At the time of reconstruction, according to the input LR image patch, the image dictionary of the class to which it belongs is found, and the dictionary atom that is closest to the image patch feature vector is found in the dictionary. The HR image patch is reconstructed according to the mapping matrix of the dictionary atom and the LR image patch. The feature clustering method is used to enhance the feature expression ability of image dictionary and improve the accuracy of the mapping matrix. Moreover, we calculate the image feature mapping matrix offline by using collaborative representation, which increases the image reconstruction speed. Experimental results show that the proposed SR method not only improves the image reconstruction efficiency, but also reconstructs more high-frequency information, making the reconstructed image closer to the real image. The proposed method can perform well both in terms of image reconstruction efficiency and image reconstruction effect, which plays an important role in the practical application of SR technology.
The Institute of Computer Applications at the School of Computer and Information, Hefei University of Technology, is mainly dedicated to the research of video image processing, machine learning and visual computing, computer graphics and visualization. Currently, there are 2 professors, 4 associate professors, and 5 lecturers. It has strong technical force and rich scientific research achievements. It has undertaken several National Natural Science Foundation, Anhui Provincial Science and Technology department and the “863” R&D project, and has published more than 200 papers in international and domestic high level journals and conferences, and received a number of national invention patents. The main research direction of Professor Wang Ronggui's team is intelligent video processing and analysis, large video data and cloud computing, intelligent video surveillance and public security. The team presided over two projects of the National Natural Science Foundation of China and one of the Natural Science Foundation of Anhui Province, and chaired a number of commissioned projects of the 38th institute of CLP Group and Keli Information Company. The research results "The key technology and application of multi-source multimode video intelligent processing" was awarded the second prize of Science and Technology Progress in Anhui Province in 2017, and the "Virtual Bayonet System" won the first prize of the 2017 CLP Group Science and Technology Award.
Wang R G, Liu L L, Yang J, et al. Image super-resolution based on clustering and collaborative representation[J]. Opto-Electronic Engineering, 2018, 45(4): 170537.