Synthesis and analysis of prediction errors and error fusion based prior for prediction algorithms

• Published in: Multimedia tools and applications 2022-06, Vol.81 (14), p.19835-19847
• Main Authors: Kumar, Neeraj, Chakravarti, Ashish Kumar, Singh, Yashveer
• Format: Article
• Language: English
• Subjects: 1205: Emerging Technologies for Information Hiding and Forensics in Multimedia Systems; Algorithms; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Error analysis; Estimation; Image reconstruction; Interpolation; Multimedia Information Systems; Pixels; Special Purpose and Application-Based Systems
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-021-11144-z

In this paper, we propose a simple and efficient post-processing algorithm to improve the accuracy of prediction based, computationally simple image reconstruction algorithms. The proposed algorithm works where there is a need to restore the missing pixels (such as interpolation, deinterlacing, sub-pixel rendering, denoising, and demosaicing). In this work, we formulate the post-processing stage as a Maximum-a-Posteriori (MAP) estimation problem. Interestingly, we find that prediction errors of a missing pixel and its neighboring pixels are also correlated, which can be utilized to improve the prediction accuracy. Therefore, we propose an efficient way of calculating prior information by estimating synthetically created prediction errors from the four connected neighbors and fuse these prediction errors based on the corresponding activity levels. Experiments demonstrate that the proposed method can significantly improve the performance of existing computationally simple prediction algorithms in terms of both objective and subjective quality with a slight increment of the computational requirement.