Image Super Resolution Reconstruction Based On Deep Convolutional Network

Image super resolution reconstruction is a technique that improves the image quality by the computer software algorithms.That is to say,a high-resolution image can be reconstructed from one or more low-resolution images of the same scene.It not only avoids the inherent limitations of the reconstructed high-resolution image by means of the hardware,such as big difficult and high cost,but also is of great theoretical and practical significance in improving medical diagnosis,image visual effects,face recognition and other aspects,which makes it have a wide range of applications in the practical problems,including criminal investigation,Satellite imaging,video surveillance and.medical image processing.In recent years,the image super-resolution reconstruction techniques have been rapidly developed and achieved promising results.With the advent of the big data era,image super-resolution reconstruction based on deep learning has become a current mainstream method.According to the current research status,the thesis mainly focuses on the following works:1.The thesis first briefly summarizes the research background and significance of image super-resolution reconstruction and its development both at home and abroad,and analyzes some conventional methods in this filed after classification and summarization of the existing super-resolution reconstruction algorithms.Then,it elaborates the related theoretical knowledge of neural network.Finally,the super-resolution reconstruction techniques we proposed are applied to the sketch face recognition problem.As to the problem,firstly,a sketch face image is transformed into a face image similar to the real ones according to the eigenface algorithm.Secondly,improve the quality of the transformed face image by image super-resolution reconstruction based on convolutional neural network.Finally,identify and analyze the face image before and after super-resolution reconstruction respectively by using linear discriminant analysis.Sketch face recognition based on super-resolution reconstruction can improve the image visual effect and increase the sketch face recognition rate effectively.2.An edge-enhanced image super resolution reconstruction algorithm based on deep network is studied in this thesis.Firstly,the low-level features of the input low-resolution image are extracted by the preprocessing network and then input it into the mixture networks.One of the mixture networks gets the advanced features through the cascade of all convolutional layers,while the other one reconstructs the edges of the the image by concatenating the convolutional network and the deconvolution network which is a mirrored version of the convolutional network.Finally,the results of the two networks are combined by the branch connection,which passed through a convolution layer to obtained the final reconstructed high-resolution image with edge enhancement.Experiment results reveal that the proposed method based on edge enhancement model achieves better reconstruction performance on quantitatively and qualitatively compared with other methods.