Flexible Reference Image Quality Assessment for Distortion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing full-reference image quality assessment (FR-IQA) methods rely on a single pristine image as a reference, ignoring the potential for multiple indistinguishable pristine-quality representations, and lack flexibility in selecting the best reference for perceptual similarity measures.
Innovation Solution
The Flexible Reference-based Equal-Quality Space (FLRE) paradigm constructs an equal-quality space using a pre-trained neural network to predict near-threshold maps, allowing for the selection of a pseudo-reference feature through a Pseudo-Reference Search (PRS) strategy, optimizing quality regression, disturbance maximization, and content losses to determine the best explanation for distorted features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single pristine image is used as reference in FR-IQA methods, then the assessment process is simple and straightforward, but the accuracy and flexibility of quality assessment deteriorates due to ignoring multiple indistinguishable pristine-quality representations
Solution Approach 1:
The patent transforms the static single-reference approach into a dynamic multi-reference system. The equal-quality space is constructed by generating multiple pristine-quality representations through near-threshold map predictions, allowing the reference to adapt and change based on the distorted image characteristics. This dynamic reference selection process improves assessment accuracy by matching the most appropriate reference from the generated space.
Solution Approach 2:
The patent changes the parameter of reference representation from a single fixed image to multiple variable representations within an equal-quality space. By varying the reference parameters (different pristine-quality representations generated through near-threshold perturbations), the system can select the optimal reference that maximizes perceptual similarity with the distorted image, thereby improving measurement precision without excessive complexity.
2Adaptability or versatility
If multiple pristine-quality representations are generated to improve assessment accuracy, then the quality assessment becomes more flexible and accurate, but the computational complexity and processing time increase
Solution Approach 1:
The patent applies preliminary action by pre-generating the equal-quality space of multiple pristine representations before actual quality assessment. The near-threshold maps are predicted and equal-quality images are synthesized in advance, creating a ready-to-use reference space. When a distorted image needs assessment, the system can quickly search and select from this pre-prepared space, reducing real-time processing time while maintaining flexibility and accuracy.
3Measurement precision
If the best reference is selected from equal-quality space using PRS strategy, then the perceptual similarity measurement is optimized, but the computational burden increases due to searching through multiple references
Solution Approach 1:
The patent applies local quality by focusing the search process on locally optimal solutions within the equal-quality space. The Pseudo-Reference Search strategy uses perceptual similarity metrics to identify the best matching reference for each specific distorted image, rather than exhaustively evaluating all possible references. This localized optimization approach improves perceptual similarity measurement accuracy while reducing overall computational power requirements.
Data Source
AI summary
A novel FR-IQA paradigm involving a flexible reference selection is proposed. It dedicates to generating the reference feature by finding the best explanation of the distorted feature among an equal-quality space constructed based on a given pristine feature. Without the ground-truth reference for distorted images with various distortion types, the quality regression loss, the disturbance maximization loss and the content loss are employed to optimize the pseudo-reference feature learning. Experimental results on five IQA benchmark databases demonstrate that combining the FLRE with the existing deep feature-based FR-IQA models can gain performance improvement.