Image Quality Assessment via Wavelet Subband Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image quality assessment methods are limited by high computational complexity, difficulty in accurately combining human visual system (HVS) bands, and inefficiencies in extracting image statistics, particularly in top-down approaches like SSIM methods that may not account for varying distortion impacts across image areas.
Innovation Solution
A method involving N-level multiresolution decomposition to separate images into approximation and detail subbands, followed by selective aggregation and weighted pooling to determine image quality, focusing on both main content and edge similarities, thereby reducing computational complexity and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional full-reference quality measures like PSNR are used, then the assessment is simple and computationally efficient, but the measure cannot sufficiently reflect human perception of image fidelity
Solution Approach 1:
The patent divides the image into multiple subbands using discrete wavelet transform (DWT), separating the image into approximation subbands and detail subbands at different decomposition levels. This segmentation allows selective processing of different frequency components, enabling perceptually accurate quality assessment by focusing on visually important regions while reducing overall computational complexity.
2Measurement precision
If multi-channel models with multiple decomposition levels are used to model HVS features, then perceptual accuracy is improved, but the device complexity and computational requirements increase significantly
Solution Approach 1:
The patent extracts only the visually important components from the full multi-level wavelet decomposition by selectively processing approximation subbands and detail subbands. Instead of analyzing all decomposition levels equally, the method extracts and weights subbands based on their perceptual importance to human vision, simplifying the overall model while maintaining perceptual accuracy.
Solution Approach 2:
The patent applies different processing and weighting strategies to different subbands based on their local characteristics and perceptual importance. Approximation subbands receive different treatment compared to detail subbands, and further differentiation is made based on orientation and frequency content, allowing the model to focus computational resources on locally important features.
3Measurement precision
If all intermediate and detail subbands are aggregated to produce edge maps, then edge detection accuracy is improved, but computational time and complexity increase
Solution Approach 1:
The patent applies partial action by selectively aggregating only certain intermediate and detail subbands to produce edge maps, rather than processing all available subbands. This selective aggregation focuses computational effort on subbands that contribute most to edge detection accuracy, reducing overall computational time while maintaining acceptable edge detection performance.
Data Source
AI summary
Method and system for determining a measure of quality for images by using multi-level decomposition are presented. Multi-level decomposition of images is performed in the wavelet domain producing subbands at each level of decomposition. Aggregation of subbands is performed across multiple levels to produce an accurate measure of image quality. By aggregating only selected subbands the computational complexity of the method is greatly reduced.