Research Of Piecewise Regression-based Single Image Super-resolution Algorithm And System

The rapid development of computer information processing technology and transmission technology accelerates the process of society informatization,which makes the digital images greatly enrich the way people learn,work and live.Image resolution is an important measure of image quality.The higher the image resolution,the richer the information it provides.Because of external imaging conditions and limitations of imaging equipment,we sometimes get poor quality,low resolution images.Image super-resolution technology can predict the corresponding high-resolution image from a low-resolution image at a low economic cost.Most of the existing super-resolution methods are learning-based single image super-resolution methods.But they have many limitations,such as high computational complexity or slow running speed when generating high-resolution images.This paper aims to explore the problem of single image super resolution using new image feature extraction methods and mapping model solving methods,which can help to improve the quality of reconstructed images while speeding up the running speed.The main work of this thesis is as follows:(1)we have proposed a Hadamard pattern-based single image super-resolution method.We perform Hadamard transform on vectorized low-resolution training data to obtain Hadamard patterns,which are used to cluster the training data.Then we calculate the mapping model from low-resolution space to high-resolution space for each class.Hadamard matrix is the operator of Hadamard transform.In the process of clustering we build a ternary decision tree.The mapping model is computed by the least square method.(2)We have proposed an extreme learning machine-based single image super-resolution method.After clustering the training data,we calculate the mapping models by extreme learning machine.(3)We have designed and implemented a single image super-resolution system,which is a practical application of our super-resolution algorithm.The experimental results show that our proposed methods can generate quality images with more high-frequency information in the fastest running speed.