Image Sparse Representation Via Convolutional Sparse Coding With Iterative Support Detection

Image Super-resolution technique reconstructs a higher-resolution image or sequence from the observed low-resolution images.As Super-resolution has been developed for more than three decades,both multi-frame and single-frame Super-resolution have significant applications in many fields,such as medical imaging,video surveillance.The high-resolution images help methods of these fields to get better recognition and classification effect.Therefore,image Super-resolution is fundamental and important in computer vision.Researchers release many methods for image super-resolution,which are generally divided into three categories:interpolation-based,reconstruction-based,and learning-based.The current best and most popular method is based on the learning method.The deep learning method for image super-resolution and the super resolution based on the ISD-CSC method released in this thesis,are all learning-based.In this thesis,based on the convolutional sparse coding method,we further research the element of"sparsity",and get the differences between the7)₁ norm and the7)₀ norm.Therefore,we import the ISD method,and release the ISD-CSC method,and apply this method in image super-resolution.The advantage of this method is that the ISD method can quickly detect the non-zero elements in the coefficient matrix and then truncate these coefficients.Then better sparse representation with better sparse coefficient matrix can be obtained,so we can get super-resolution image with higher signal-to-noise ratio and better visual effect.