Multi-measures fusion based on multi-objective genetic programming for full-reference image quality assessment

TL;DR

This paper introduces a novel multi-measures fusion technique for image quality assessment that automatically selects and weights measures using multi-objective genetic programming, outperforming existing methods.

Contribution

It proposes a new fusion method leveraging multi-objective genetic programming to optimize measure selection and weighting for improved image quality assessment.

Findings

Outperforms state-of-the-art IQA fusion methods

Effective automatic measure selection and weighting

Validated on large public image databases

Abstract

In this paper, we exploit the flexibility of multi-objective fitness functions, and the efficiency of the model structure selection ability of a standard genetic programming (GP) with the parameter estimation power of classical regression via multi-gene genetic programming (MGGP), to propose a new fusion technique for image quality assessment (IQA) that is called Multi-measures Fusion based on Multi-Objective Genetic Programming (MFMOGP). This technique can automatically select the most significant suitable measures, from 16 full-reference IQA measures, used in aggregation and finds weights in a weighted sum of their outputs while simultaneously optimizing for both accuracy and complexity. The obtained well-performing fusion of IQA measures are evaluated on four largest publicly available image databases and compared against state-of-the-art full-reference IQA approaches. Results of comparison reveal that the proposed approach outperforms other state-of-the-art recently developed fusion approaches.