Neural Image Filter Selection for Low-Loss Artifact Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques are inefficient, resource-intensive, and often result in sub-optimal image quality due to the use of multiple neural networks and a lack of focus on overall image field of view, leading to visual artifacts and poor immersive experiences.
Innovation Solution
A computationally-efficient method using an analysis neural network to select and apply image processing filters with minimal loss for different parts of an image, improving visual quality by identifying and correcting defects like noise and blur.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple different neural networks are employed to correct different artifacts, then comprehensive artifact correction is achieved, but training complexity and processing resource consumption increase significantly
Solution Approach 1:
The patent combines multiple specialized neural networks into a single unified neural network that can handle multiple types of image artifacts simultaneously. This unified network integrates the functionality of what would otherwise be separate networks for different artifact types, reducing training complexity while maintaining comprehensive correction capabilities through a single integrated training process
2Reliability
If multiple neural networks are used in cascade manner, then diverse artifacts can be corrected, but processing time increases and frame rate decreases
Solution Approach 1:
The patent merges multiple cascade neural networks into a single unified network that processes all artifact corrections in one pass rather than requiring sequential processing through multiple networks. This eliminates the time penalty of cascaded processing while maintaining the ability to correct diverse artifacts through integrated multi-task learning
3Manufacturing precision
If existing image processing techniques focus only on gaze region or peripheral region, then specific region quality is improved, but overall image field of view quality suffers with visual artifacts
Solution Approach 1:
The patent creates a universal neural network that simultaneously optimizes image quality across all regions of the field of view - gaze region, peripheral region, and intermediate regions. The unified network applies learned corrections globally rather than region-specifically, ensuring consistent quality throughout the entire image while maintaining the ability to adapt to different artifact types in different areas
4Reliability
If multiple neural networks are employed, then comprehensive artifact correction is achieved, but computational resources are wasted when not all artifacts are present in every image
Solution Approach 1:
The patent implements a unified neural network that performs partial corrections only where needed rather than applying all correction operations uniformly across the entire image. The network learns to identify and correct only the artifacts actually present in each image region, avoiding unnecessary computational expenditure on regions or artifact types that are not problematic
Data Source
AI summary
A computer-implemented method includes utilising an analysis neural network to select a first image processing filter from amongst a plurality of image processing filters that is to be applied to a given part of an input image, wherein the analysis neural network is trained to select an image processing filter having a minimum loss for the given part of the input image as the first image processing filter, wherein respective first image processing filters are selected for different parts of the input image; and applying the respective first image processing filters to the different parts of the input image.