Adaptive Window Stereo Matching for 3D Face Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In three-dimensional face measurement, there is a demand for accurately reproducing feature portions like eyes, nose, and mouth while smoothly reproducing the surface of areas like cheeks, which existing stereo methods struggle to achieve effectively, especially when dealing with movements and avoiding intense light on the eyes.
Innovation Solution
An image processing apparatus that sets windows on base and reference images, extracts image feature values, and adapts the window size and shape based on these values to enhance correspondence and three-dimensional coordinate calculation, using phase-only correlation for robust and accurate stereo matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fixed-size window is used for stereo correspondence search, then the processing is simple and fast, but the measurement precision deteriorates because it cannot adapt to different regions (feature portions vs. flat surfaces)
Solution Approach 1:
The patent applies dynamics by making the window size adaptive rather than fixed. The window size is dynamically adjusted based on the standard deviation of grayscale values in the reference image - regions with high variance (feature portions like eyes, nose, mouth) receive smaller windows, while regions with low variance (flat surfaces like cheeks) receive larger windows. This dynamic adaptation resolves the contradiction by improving measurement precision in different regions without requiring complex manual configuration.
Solution Approach 2:
The patent changes the window size parameter based on image content characteristics. By calculating the standard deviation of grayscale values and comparing it to a threshold, the system automatically adjusts the window size parameter to match the local image features. This parameter change strategy enables precise 3D measurement of both detailed feature portions and smooth surfaces, resolving the contradiction between measurement precision and processing simplicity.
2Measurement precision
If a small window size is used for feature portions, then the measurement precision improves, but the noise resistance deteriorates
Solution Approach 1:
The patent applies local quality by assigning different window sizes to different regions of the image based on their local characteristics. Feature portions (eyes, nose, mouth) with high grayscale variance receive smaller windows for precise localization, while flat surfaces (cheeks) with low variance receive larger windows for better noise resistance. This local differentiation resolves the contradiction by optimizing the window size for each region's specific requirements rather than using a uniform size throughout.
Solution Approach 2:
The system changes the window size parameter dynamically based on the calculated standard deviation of grayscale values in each region. When the standard deviation exceeds a threshold (indicating a feature portion), a smaller window size is selected to improve precision. When the standard deviation is below the threshold (indicating a flat surface), a larger window size is selected to improve noise resistance. This adaptive parameter change resolves the contradiction between precision and reliability.
3Reliability
If a large window size is used for flat surfaces, then the noise resistance improves, but the measurement precision deteriorates due to loss of detail
Solution Approach 1:
The patent applies local quality by differentiating window sizes according to local image characteristics. Flat surface regions (cheeks) with low grayscale variance are assigned larger windows that provide noise resistance while still capturing sufficient surface detail. Feature portions (eyes, nose, mouth) with high variance are assigned smaller windows to preserve fine details. This local differentiation resolves the contradiction by matching window size to the specific requirements of each region.
Solution Approach 2:
The system changes the window size parameter based on the standard deviation of grayscale values in each region. For flat surfaces with low standard deviation, larger window sizes are applied to improve noise resistance while maintaining adequate surface detail. For feature portions with high standard deviation, smaller window sizes are applied to preserve fine details. This adaptive parameter adjustment resolves the contradiction between noise resistance and detail precision.
Data Source
AI summary
An apparatus received input of a base image and a reference image that are images picked up by two cameras provided on left and right. A window setting portion determines a size of a window based on an edge value of an image around a gaze point in the base image. A position deviation operation portion operates deviation between the image in the window of the base image and the image in the window of the reference image. A three-dimensional coordinate operation portion operates a three-dimensional coordinate of a surface of an object based on deviation between the images.