Convolutional Color Correction Using Log-Chromaticity Histograms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital imaging technologies face challenges in accurately inferring and correcting the color of illumination in images, known as the color constancy problem, which is under-constrained and affects the accuracy of white-balance adjustments across different image sensors.
Innovation Solution
A color constancy technique is developed using convolutional neural networks and structured prediction, framing the problem as a discriminative machine learning issue by analyzing the scaling of color channels and translating log-chromaticity histograms to estimate illumination, thereby reducing error rates by up to 40% on standard benchmarks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional color constancy techniques are used, then white-balance adjustment can be performed, but the accuracy is insufficient and error rates remain high
Solution Approach 1:
The patent transforms the color constancy problem from direct illuminant estimation to histogram translation in log-chromaticity space. By changing the parameter space from raw color values to log-chromaticity coordinates, the method achieves more accurate and reliable white-balance correction, reducing error rates by up to 40% on standard benchmarks.
Solution Approach 2:
The patent replaces traditional mechanical or algorithmic white-balance adjustment methods with a convolutional neural network-based discriminative learning system. This substitution enables the system to learn complex patterns in chrominance histograms and automatically determine optimal white-balance corrections, significantly improving accuracy over conventional techniques.
2Measurement precision
If color correction is performed to achieve accurate white-balance, then color accuracy improves, but the solution becomes dependent on specific image sensor characteristics
Solution Approach 1:
The patent develops a universal color constancy technique that works across different image sensor types. By training the convolutional neural network on diverse training images from multiple sensors and using sensor-independent features (chrominance histograms in log-space), the system achieves accurate white-balance correction that is not dependent on specific sensor characteristics, thereby improving adaptability while maintaining color accuracy.
3Measurement precision
If discriminative machine learning with convolutional neural networks is used, then error rates reduce by up to 40%, but computational complexity and processing requirements increase
Solution Approach 1:
The patent segments the color constancy problem into distinct processing stages: (1) computing chrominance histograms from input image, (2) convolving histograms with learned filters to produce heat maps, (3) selecting optimal white-balance parameters from heat map maxima, and (4) applying correction to the image. This segmentation allows the complex discriminative learning task to be broken into manageable steps, reducing overall computational complexity while maintaining high accuracy.
Solution Approach 2:
The patent performs preliminary training of the convolutional neural network offline using large datasets of labeled images. The learned filters and parameters are stored and reused during actual white-balance correction operations. This preliminary action separates the computationally intensive learning phase from the real-time correction phase, reducing the computational burden during actual image processing while maintaining high accuracy.
Data Source
AI summary
A computing device may obtain an input image. The input image may have a white point represented by chrominance values that define white color in the input image. Possibly based on colors of the input image, the computing device may generate a two-dimensional chrominance histogram of the input image. The computing device may convolve the two-dimensional chrominance histogram with a filter to create a two-dimensional heat map. Entries in the two-dimensional heat map may represent respective estimates of how close respective tints corresponding to the respective entries are to the white point of the input image. The computing device may select an entry in the two-dimensional heat map that represents a particular value that is within a threshold of a maximum value in the heat map, and based on the selected entry, tint the input image to form an output image.