Image Processing Apparatus Using Neural Network Feature Integration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques face challenges in generating high-quality images from multiple low-quality images with different attributes, as they often require pixel-level alignment and struggle to effectively address systematic degradations like chromatic aberration and spherical aberration through image averaging.
Innovation Solution
An image processing apparatus that uses a combination of neural networks to derive and integrate features from multiple images, allowing for the generation of high-quality images by transforming and integrating these features rather than aligning pixels, thereby improving image quality through a process that includes feature connection and deconvolution operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If images undergo projective transformation and pixel-level superimposition, then image alignment is achieved, but processing complexity increases and systematic degradations cannot be effectively removed
Solution Approach 1:
The patent segments the image processing task into feature extraction, feature integration, and image generation stages. Instead of performing complex pixel-level alignment operations, the system extracts features from multiple images, integrates these features, and then generates the final aligned image, thereby simplifying the processing while maintaining alignment precision.
Solution Approach 2:
The patent replaces the mechanical pixel-level transformation and superimposition system with a feature-based processing system using neural networks. This substitution eliminates the need for complex projective transformations and direct pixel manipulation, reducing processing complexity while achieving the same alignment objective.
2Reliability
If image averaging is used to improve image quality, then noise reduction is achieved, but systematic degradations like chromatic aberration and spherical aberration cannot be removed
Solution Approach 1:
The patent extracts features from multiple images that represent the underlying scene structure, separating these meaningful features from systematic degradations such as chromatic aberration and spherical aberration. By integrating only the extracted features rather than the complete images, the system eliminates systematic degradations while preserving useful information.
Solution Approach 2:
The patent introduces feature integration as an intermediary step between image acquisition and final image generation. This intermediary process allows the system to combine information from multiple images while filtering out systematic degradations, achieving both noise reduction and removal of aberrations.
3Manufacturing precision
If multiple images with different attributes are processed, then image quality improvement is achieved, but processing time and computational load increase
Solution Approach 1:
The patent performs feature extraction as a preliminary action before integration. By pre-processing the images to extract only the essential features needed for quality improvement, the system reduces the computational load and processing time required for subsequent integration and generation steps.
Solution Approach 2:
The patent changes the parameter representation from pixel-level data to feature-level data. This parameter transformation reduces the dimensionality and complexity of the processing task, enabling faster integration of multiple images while maintaining or improving image quality.
Data Source
AI summary
An image processing apparatus includes: an acquisition unit configured to acquire a plurality of images each capturing an identical target and having a different attribute; a derivation unit configured to derive features from the plurality of images, using a first neural network; an integration unit configured to integrate the features derived from the plurality of images; and a generation unit configured to generate a higher quality image than the plurality of images from the feature integrated by the integration unit, using a second neural network.