Neuro-Template Image Correction for Faster AWB Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic white balancing (AWB) techniques are memory and time inefficient due to nearest neighbor matching and lack of attribute information usage, leading to scalability issues and suboptimal image correction.
Innovation Solution
A neuro-template based method using an AI module to generate template images through color, attribute, and shifted pixel transformations, assigning weights to these images for correction, and applying them via matrix multiplication to achieve efficient image correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If nearest neighbor matching is performed on each pixel to perform AWB, then AWB correction is achieved, but processing time increases and scalability deteriorates
Solution Approach 1:
The patent divides the image processing task into two stages: (1) pre-compute transformation matrices for a limited set of basis images representing different lighting conditions, and (2) represent each pixel as a linear combination of these basis images using the pre-computed matrices. This segmentation allows avoiding exhaustive nearest neighbor searches while maintaining correction accuracy.
Solution Approach 2:
The patent performs preliminary computation of transformation matrices during an offline training phase using a dataset of images with known ground truth corrections. These pre-computed matrices are then reused during actual AWB processing, eliminating the need for time-consuming real-time computations and enabling fast inference.
2Measurement precision
If nearest neighbor matching is performed with a large number of neighboring pixels, then AWB correction quality improves, but memory requirements increase
Solution Approach 1:
The patent creates a compact representation of the image by expressing each pixel as a linear combination of a small number of basis images (typically 3-10). This copied representation uses significantly less memory than storing and processing large numbers of neighboring pixels, while still capturing the essential color information needed for accurate AWB correction.
3Device complexity
If attribute information such as race, hair texture, and background categories is not used, then processing complexity is reduced, but image correction accuracy deteriorates
Solution Approach 1:
The patent transforms the image data into a different parameter space by applying color kernel transforms that operate on color channels (e.g., converting RGB to a different color basis). This parameter transformation allows the model to capture attribute information implicitly through color relationships without requiring explicit detection and processing of attributes like race, hair texture, or background categories.
Data Source
AI summary
A device and method for performing image correction may include: receiving an input image to be corrected; generating a plurality of template images based on the input image, wherein each template image from among the plurality of template images is generated based on at least one of an image feature transformation; providing the plurality of template images to an artificial intelligence (AI) module; assigning, using the AI module, weights from among zero and non-zero magnitudes to each template image based on an extent and a type of correction to be applied to each template image; and applying the assigned weights to the plurality of template images to obtain a corrected output image.