| Autor: |
S., Greeshma M., R., Bindu V. |
| Zdroj: |
Multimedia Tools & Applications; Jan2024, Vol. 83 Issue 1, p253-278, 26p |
| Abstrakt: |
Deep networks have emerged as a dominant solution in many research areas recently. Numerous approaches based on deep networks have been developed for image super-resolution problems with good performance. The Super-resolution Convolutional Neural Network model attempts to direct feature learning from low to high-resolution images but sometimes fails in network training when noisy examples are presented. In this work, we propose a super-resolution model that exploits an image represented by the deep structures while characterized by the primitive prior. More specifically, we discuss the use of traditional sparse representation being still sensible and combine processing of primitive prior with deep learning structure to attain further enhanced results. In addition, we apply the sparsity of primitive prior to the super-resolution problem and generate a sharp-edged high-resolution image from low-resolution image. The primitive priors are more informative structures, which can retain the high-resolution image effectively. The network simulation based on primitive prior leads to more effective network training and fine-tuning of the network. On evaluating the super-resolution model on different low-resolution images, an enhanced performance over existing algorithms is achieved in quantitative and qualitative validations. The quantitative results show that the proposed network outshines the other state-of-the-art approaches, surpassing the metric values. Qualitative analysis using the primitive network model achieves good visual quality concerning ridge and corner details. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink