Abstract

The super resolution of a sole image is a fundamental challenge in low-level computer vision with many applications. The intention of Single Image Super Resolution (SISR) is to transform low-resolution (LR) images into high-resolution (HR) images with all the necessary edge structures and texture information. The HR images provide more details that can be used for various purposes such as security, medical imaging, etc. However, the reconstructed image has some loss in detail parts present in the image, and attaining better accuracy with less error is quite challenging. To address these challenges, a regression network based super resolution (RNSR) is developed that converts LR images into HR images. A deep regression network with 50 layers is designed for the SISR. According to the simulation analysis, the proposed RNSR method achieves 98% accuracy, 0.02% error, 97% precision, and 94% specificity for converting LR multimedia text images to HR images. Based on the performance of the proposed RNSR method, the regression network can generate high quality images.