Intrinsic Image Generation for Illumination-Invariant Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer systems for object recognition in images face challenges in varying illumination conditions, which complicate the identification of objects due to the need to distinguish between material and illumination effects, leading to inefficiencies and inaccuracies in image analysis.
Innovation Solution
The method involves generating intrinsic images that separate material and illumination aspects of a standard image, using techniques such as bi-illuminant dichromatic reflection models and chromaticity planes, to create a classifier that can identify objects of interest by isolating material features from illumination effects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If differential or frequency based features of grayscale imagery are used to achieve illumination and color invariance, then the system can function under varying illumination conditions, but the complexity of image analysis increases and affects efficiency
Solution Approach 1:
The patent segments the image analysis process into distinct stages: preprocessing to generate illumination-invariant features, followed by classification. By separating the feature extraction stage from the classification stage, and by creating simplified invariant features through preprocessing, the system achieves illumination invariance without carrying forward the full complexity of the original image data through the entire analysis pipeline.
Solution Approach 2:
The patent applies preliminary processing to the images before classification, specifically transforming images into illumination-invariant representations through operations like gradient calculation and normalization. This preliminary action removes illumination variations before the main classification task, thereby reducing the complexity that would otherwise be present during the primary analysis phase.
2Measurement precision
If conventional image analysis methods are used without separating material and illumination effects, then the system operates simpler, but accuracy decreases under varying illumination conditions
Solution Approach 1:
The patent extracts and removes the illumination component from the image data through preprocessing operations. By taking out the illumination effects and creating illumination-invariant features, the system isolates the material properties of objects, thereby improving recognition accuracy without requiring the full complexity of simultaneous material and illumination analysis.
Solution Approach 2:
The patent transforms the image parameters through mathematical operations such as gradient calculation, normalization, and thresholding to create illumination-invariant features. These parameter changes convert the original illumination-dependent image data into a representation that is insensitive to illumination variations, improving accuracy while maintaining a manageable level of processing complexity.
Data Source
AI summary
In a first exemplary embodiment of the present invention, an automated, computerized method for learning object recognition in an image is provided. According to a feature of the present invention, the method comprises the steps of providing a training set of standard images, calculating intrinsic images corresponding to the standard images and building a classifier as a function of the intrinsic images.