Video Super Resolution and Performance Enhancement of Mixture Mapping Model by Deep Learning De-Noising
Abstract
Super resolution is method of reconstruction of high resolution image or video from low resolution images or frames. This paper present the mixture mapping method of super resolution by adapting deep convolutional neural network based de-noising method. In de-noising method blind Gaussian noise is removed by feed forward method and batch stabilization is used to stabilize the residual image or frame. The super resolution technique is designed by separating less information and more information patch features by curvature difference method and by mixture mapping technique high resolution patches are reconstructed. The algorithm is implemented in MATLAB 2018 and quality of image is estimated with design metrics like Peak signal to noise ratio (PSNR), Structural Similarity index method (SSIM), IFC and FSIM. Experimental results shows that by adding deep CNN de-noising, quality of mixture mapping model is improved as compare to previously published methods like Bicubic, ScSR, SRCNN, SelfExSR, MMPM-G and MMPM-S. Performance parameters of proposed methods for image dataset are as PSNR is 38.20dB, SSIM is 0.9748, FSIM is 0.9824 and IFC is 8.8698.