CCOCSA-based multi-frame sparse coding super-resolution via mutual information-based weighted image fusion

Article Type

Research Article

Publication Title

Multimedia Tools and Applications

Abstract

Image super-resolution (SR) is one of the most urgent requirements in many applications in computer vision. Though many techniques have been proposed to date, they are not suitable in real-life applications because of their lack of performance when the low resolution (LR) images are noisy. In the single-frame SR mechanism, the output image can not reproduce the information which is lost due to sub-sampling at the time of image formation. Multi-frame image SR technique can solve this problem to a reasonable extent. However, most of the multi-frame image SR techniques are based on image registration which is noise sensitive. The registration errors in noisy LR images deteriorate the performance of the registration process and hence the performance of the multi-frame SR algorithms. This paper addresses this problem and proposes a novel mutual information-based multi-frame sparse coding super-resolution via Chaotic Centroid-Oppositional Crow Search Optimization Algorithm (CCOCSA)-based image registration. The proposed SR algorithm is accomplished in two steps: (1) CCOCSA-based image registration and (2) Mutual information-based weighted mean multi-frame super-resolution reconstruction through sparse representation. The proposed CCOCSA is a noise-robust optimization that provides better accuracy in estimating registration parameters. We achieve this by enhancing the exploration by dynamically varying the control parameters and introducing chaos within the algorithm. The weighted mean of the registered LR frames based on mutual information provides a good noise reduction capability. Also, it gathers the information from all the LR frames on the registered weighted mean frame, which enriches the information in the reconstructed SR image. The performance of the proposed algorithm is tested through the experiments on four benchmark datasets (Vid4, McMaster, Milanfar, and Kodak) for videos and still images. We verify the performance of the proposed technique by computing four well-known quality metrics, such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), Edge Keeping Index (EKI), blur-metric, and Perceptual SIMilarity (PSIM) measure. The experimental results show that the proposed SR algorithm outperforms the others quantitatively and qualitatively.

First Page

2427

Last Page

2471

DOI

10.1007/s11042-023-15647-9

Publication Date

1-1-2024

Link to Full Text

Share

COinS