Convolutional Color Correction via Chrominance Histogram Heat Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current color constancy techniques face challenges in accurately inferring and correcting the illumination of digital images, as they are under-constrained and often result in high error rates, especially when distinguishing between object colors and illuminant colors.
Innovation Solution
A color constancy technique is developed using discriminative machine learning with convolutional neural networks, framing the problem as a structured prediction task. This involves generating a two-dimensional chrominance histogram from an input image, convolving it with a filter to create a heat map, and selecting the entry closest to the white point to tint the image, thereby addressing the under-constrained nature of color constancy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional color constancy techniques are used to infer and correct illumination color, then the process is simpler, but the error rate increases significantly
Solution Approach 1:
The patent transforms the color constancy problem from direct pixel value analysis to histogram space analysis. By converting image colors to chrominance histograms and then to log-chromaticity histograms, the method operates in a transformed dimensional space where illumination effects manifest as predictable translations. This dimensional transformation enables the use of convolution operations to model illumination variations, significantly improving accuracy over traditional methods that work directly in color space.
Solution Approach 2:
The patent introduces chrominance histograms as an intermediary representation between the raw image data and the final white balance correction. Instead of directly analyzing pixel colors to determine illumination, the method first computes chrominance histograms that capture the distribution of colors in the image. This intermediary step filters out noise and provides a more robust representation that is less sensitive to individual pixel variations, thereby improving the reliability of illumination estimation.
2Measurement precision
If hardware-specific algorithms are used for color correction, then the accuracy for that specific sensor is improved, but the adaptability to different image sensors decreases
Solution Approach 1:
The patent develops a universal color constancy algorithm that operates on chrominance histograms rather than sensor-specific pixel data. The method uses convolution operations with learnable filters that capture general patterns of illumination variation across different sensors and conditions. By training on diverse datasets from multiple sensors and applying the same histogram-based approach, the system achieves cross-sensor generalization while maintaining high accuracy on individual sensors, making the algorithm universally applicable.
3Adaptability or versatility
If the color constancy problem is treated as an under-constrained inference problem, then the theoretical flexibility is maintained, but the practical error rate increases
Solution Approach 1:
The patent replaces traditional mechanical inference approaches (direct optimization of likelihood functions) with a convolution-based signal processing approach. Instead of attempting to solve the under-constrained problem through complex probabilistic inference, the method transforms the problem into histogram space where illumination effects become additive translations. Convolution with learned filters then efficiently finds the optimal translation, providing a computationally tractable solution that maintains theoretical flexibility while achieving practical accuracy.
Data Source
Figure 1
Figure 2
Figure 3
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.