Auto White Balance Temporal Stabilization via Confidence Locking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Auto White Balance (AWB) algorithms face temporal instabilities and incorrect white point estimations, especially when dealing with difficult scenes, leading to unnatural color casts and user experience degradation, due to reliance on smoothing filters and computationally expensive high-frequency feature detection.
Innovation Solution
A method that locks or unlocks white point information based on confidence estimates, using a histogram to compute confidence values and determine threshold values, allowing the white point to remain unchanged during unreliable estimates and adapt when conditions improve, without requiring external inputs or dedicated hardware.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If a smoothing filter is used to stabilize AWB output, then temporal stability of white point is improved, but the system converges to wrong white point estimates when camera is pointed at difficult scenes
Solution Approach 1:
The patent applies dynamics by making the smoothing filter adaptive rather than static. The filter's behavior changes dynamically based on scene analysis: it applies strong smoothing when scenes are stable and relaxed when scenes change or contain difficult content. This allows the system to maintain temporal stability during normal operation while avoiding convergence to wrong white points when conditions change, resolving the contradiction between stability and accuracy.
Solution Approach 2:
The patent implements feedback by continuously analyzing scene content (color histograms, feature detection) and using this information to adjust the smoothing filter's strength. The system monitors whether the current scene is stable or changing, and feeds this information back to control the filter's convergence behavior. This feedback mechanism prevents the filter from locking onto incorrect white points in difficult scenes while maintaining stability in normal conditions.
2Stability of the object's composition
If convergence time T is increased to improve stabilization, then temporal stability is improved, but the system takes longer to adapt to scene changes and light source changes
Solution Approach 1:
The patent makes the convergence time T dynamic rather than fixed. The system adjusts T based on detected scene conditions: using longer convergence times for stability during normal operation, and shorter convergence times when scene changes or light source changes are detected. This dynamic adjustment allows the system to achieve both long-term stabilization and rapid adaptation to changes, resolving the contradiction between stability and adaptability.
Solution Approach 2:
The system uses feedback from scene analysis (detecting scene changes, light source changes, and difficult content) to adjust the convergence time parameter. When changes are detected, the feedback mechanism reduces T to enable faster adaptation; when scenes are stable, it increases T for better stabilization. This feedback-controlled parameter adjustment resolves the contradiction between stabilization and response speed.
3Measurement precision
If high-frequency image feature detection and matching is used for scene change detection, then detection accuracy is improved, but computational cost and hardware requirements increase
Solution Approach 1:
The patent segments the scene analysis task into multiple levels: using computationally efficient color histogram analysis as the primary method for detecting scene and light source changes, and reserving high-frequency feature detection for specific cases where higher accuracy is needed. This segmentation allows the system to achieve sufficient detection accuracy while minimizing computational cost and hardware requirements for most operating conditions.
Solution Approach 2:
The system changes parameters (such as the threshold for triggering high-frequency feature detection) based on operating conditions. Instead of always using computationally expensive high-frequency feature detection, the system adjusts its analysis depth: using simpler color-based detection when conditions permit, and only invoking complex feature matching when necessary. This parameter-based adaptation resolves the contradiction between detection accuracy and computational complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer implemented method for temporally stabilizing White Point (WP) information in an Auto White Balance (AWB) process. The method comprises: receiving lux level, AWB statistics, and input WP; generating a histogram of the AWB statistics; generating, from the histogram, a confidence value for the input WP; determining a major colour count; determining a threshold value based on the major colour count and the lux level; setting an update state to locked when the confidence value is below the threshold value and otherwise setting it to unlocked; final WP information is obtained from a previous frame when the update state is locked and otherwise from the input WP information. During the lock state, the WP is kept unchanged for as long as the AWB algorithm is in a difficult scene. Once the difficult scene has passed, the update parameter is set to unlock such that the WP may again change.