Human Visual System Modeling for Perceptual Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods using low-level metrics like mean squared error often result in blurry images that are not perceptually pleasing, and models of the human visual system that are not differentiable require manual tuning, lacking generalization.
Innovation Solution
A computer-implemented method using two artificial neural networks trained with differentiable functions to model the human eye and visual cortex, enabling end-to-end training on image data to improve perceptual quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If mean squared error is used for image processing, then computational simplicity is improved, but image perceptual quality deteriorates
Solution Approach 1:
The patent introduces an intermediary computational framework that bridges simple pixel-level operations and complex perceptual modeling. By decomposing the image processing into multiple stages (pixel-level operations, patch-level operations, and global operations), each stage can use appropriately complex methods without overwhelming computational cost, thereby achieving both simplicity and high perceptual quality.
Solution Approach 2:
The image processing method is segmented into three distinct operational levels: pixel-level operations for basic processing, patch-level operations for local structure preservation, and global operations for overall image quality. This segmentation allows each level to optimize for its specific function while maintaining computational efficiency across the entire processing pipeline.
2Measurement precision
If non-differentiable visual system models are used, then modeling accuracy of human vision is improved, but model training flexibility deteriorates
Solution Approach 1:
The patent replaces traditional mechanical/manual parameter tuning with a learned neural network system. By formulating the visual system model as a differentiable neural network, the system automatically learns optimal parameters from data through gradient-based optimization, eliminating the need for manual experimentation while maintaining high modeling accuracy of human visual perception.
3Adaptability or versatility
If manual parameter tuning is used for visual system models, then model adaptation to specific tasks is improved, but time consumption and labor deteriorates
Solution Approach 1:
The patent implements self-service by enabling the visual system model to automatically adapt to specific tasks through learned parameters. The differentiable neural network performs self-adjustment during training by optimizing its internal parameters via backpropagation, eliminating the need for external manual tuning and significantly reducing both time consumption and human labor while maintaining task-specific adaptability.
Data Source
AI summary
A computer-implemented method of processing image data using a model of the human visual system. The model comprises a first artificial neural network system trained to generate the first output data using one or more differentiable functions configured to model the generation of signals from images by the human eye, and a second artificial neural network system trained to generate the second output data using one or more differentiable functions configured to model the processing of signals from the human eye by the human visual cortex. The method comprises receiving image data representing one or more images, processing the received image data using the first artificial neural network system to generate first output data, processing the first output data using a second artificial neural network system to generate second output data. Model output data is determined from the second output data, and output for use in an image processing process.