Study On Image Super Resolution Reconstruction Algorithm

Image resolution is an important measure of image quality.In the process of image generation,the resolution of image is reduced due to hardware limitations of imaging system,environmental interference,object movement and so on.Image super-resolution reconstruction algorithm is designed to obtain high resolution images from existing low resolution images.The work of this thesis is based on studying and summarizing the relevant theories,classical algorithms and the latest developments of existing super-resolution technology for the problems in practical applications.These problems include edge saw tooth,process efficiency and motion estimation error issues.The major research results are as follows:1.Aiming at the edge serrated problem,an edge adaptive image interpolation algorithm based on Bandelet is proposed.The image features extracted by Bandelet transform can be divided into edge and no edge quad tree block: for the edge blocks,non-uniform linear interpolation method based on orientation is used and for the no edged blocks,bilinear interpolation method is used.Finally,we get the high resolution image whose edges have been improved.Furthermore,the image quad tree only needs to be constructed one times.The image quad tree need not be reconstructed and it greatly improves the efficiency.2.Aiming at the inaccurate motion estimation problem,a super-resolution reconstruction algorithm for sequence images based on perspective projection transform is proposed.In this algorithm,the SIFT algorithm is first used to extract the feature points of the reference image as well as moving images and find initial match points by Euclid distance,and then the error match points are removed by ratio method.Finally,at least four-pair match points are taken into the three-dimensional Homography perspective projection mode to solve the motion matrix.Experiments show that when the true motion is of affine transformation feature,image super-resolution reconstruction result is equal for affine transformation model and perspective projection transformation model;when the true motion is of perspective projection transformation feature,the latter is better than the former.3.Aiming at the application cases,a GUI software of image super-resolution reconstruction system based on Matlab is designed.Image super-resolution reconstruction is usually a system engineering,different scenarios need to use different processes.By Matlab GUI programming,a flexible software platform for image and video processing is designed.Such as interception of processed fragments in temporal and spatial,converting video into picture formats,de-noising,de-blurring,contrast enhancement,image super-resolution reconstruction etc.In order to facilitate the addition of the new algorithm,a graphical interface is reserved.