Picture-Wise JND Prediction Through Perceptual Distortion Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing JND models struggle to accurately estimate the JND threshold for entire images, particularly for distorted images, as they primarily focus on local pixel or frequency-based estimations and fail to account for image quality levels.
Innovation Solution
A multi-class perceptual distortion discriminator is trained to conduct perceptual distortion discrimination on raw and compressed images, followed by preset image-level JND search strategies for fault tolerance, predicting the image-level JND threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixel domain-based or frequency domain-based local JND threshold estimation models are used, then the calculation of local JND thresholds for individual pixels or frequencies is achieved, but the accurate estimation of JND threshold for the entire image is difficult
Solution Approach 1:
The patent transitions from local pixel/frequency domain estimation to image-level quality assessment by introducing a new dimensional perspective - evaluating the entire image as a unified entity rather than summing local measurements. The perceptual distortion discriminator operates on image-level features to predict overall quality, fundamentally changing the assessment dimension from local to global.
Solution Approach 2:
The perceptual distortion discriminator is designed as a universal model that can assess images at multiple quality levels and handle both raw and distorted images. This single model replaces the need for separate local JND estimation procedures, providing multi-functional capability for comprehensive image quality evaluation.
2Adaptability or versatility
If traditional JND models focus on raw image estimation, then the JND threshold calculation for undistorted images is achieved, but the application to distorted images is restricted
Solution Approach 1:
The patent creates a dynamic quality assessment system where the perceptual distortion discriminator adapts its predictions based on the input image's quality level. The model dynamically adjusts its evaluation criteria to account for different distortion scenarios, enabling versatile application across various image conditions while maintaining prediction accuracy through quality-aware processing.
Solution Approach 2:
The system changes the assessment parameters based on image quality level. Instead of using fixed JND thresholds, the perceptual distortion discriminator adjusts its prediction parameters according to the perceived quality level of the input image, enabling accurate assessment across different distortion conditions by dynamically modifying evaluation criteria.
3Measurement precision
If image-level JND prediction is implemented, then the overall image quality assessment is improved, but the prediction deviation needs to be reduced for better accuracy
Solution Approach 1:
The patent incorporates feedback mechanisms where the perceptual distortion discriminator's predictions are refined through iterative optimization. The model learns from prediction outcomes and adjusts its parameters to minimize deviation, creating a feedback loop that continuously improves prediction accuracy and consistency for image-level JND assessment.
Solution Approach 2:
The system performs preliminary quality assessment through the perceptual distortion discriminator before final JND threshold determination. This preliminary evaluation prepares the prediction by establishing initial quality level estimates, which are then refined to reduce deviation and improve the reliability of the final image-level JND prediction.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
A prediction method, device, equipment, and storage medium for the image-level JND threshold, comprising: perceptual distortion discrimination is conducted on the raw image and on the compressed images in the compressed image set of the said image through trained multi-class perceptual distortion discriminator to obtain the set of perceptual distortion discrimination results (S 101), and preset image-level JND search strategies are adopted for fault tolerance of the said set of perceptual distortion discrimination results to predict the image-level JND threshold of the said image (S102), thus reducing the prediction deviation of the image-level JND threshold, improving the prediction accuracy of the image-level JND threshold, and bringing the predicted JND threshold closer to the human visual system's perception of the quality of the entire image.