Red-Eye Detection Using Size-Segmented Multi-Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Reliable red-eye detection in digital images is challenging due to high false positive rates, especially in complex images, and the computational efficiency of artificial neural networks (ANNs) is impaired when dealing with varied and complex input images.
Innovation Solution
A multi-classifier system is employed to identify and eliminate false red-eye objects by using a combination of feature extractors and trained classifiers optimized for different object sizes, along with a simple classifier and filters to refine the detection, and features derived from red, green, and blue intensity values are used to classify objects as red-eye or non-red-eye.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single artificial neural network classifier is used for red-eye detection, then the system is simple to implement, but the classification accuracy deteriorates when dealing with complex images and varied object sizes
Solution Approach 1:
The patent divides the red-eye detection system into multiple specialized classifiers, each trained to handle specific object size ranges (small, medium, large). This segmentation allows each classifier to focus on particular features and patterns relevant to its size category, improving overall classification accuracy without requiring a single complex universal classifier.
Solution Approach 2:
The system dynamically selects and applies different classifiers based on the size of the detected red-eye candidate objects. The classifier selection is adaptive and depends on the input image characteristics and object dimensions, allowing the system to optimize performance for each specific detection scenario rather than using a fixed single classifier.
2Reliability
If a complex artificial neural network is used to improve classification accuracy, then detection reliability improves, but computational efficiency deteriorates
Solution Approach 1:
The computational load is segmented across multiple smaller, specialized classifiers rather than one large complex network. Each classifier handles a specific subset of objects by size, reducing the computational complexity required for each individual classification task while maintaining high overall detection reliability.
Solution Approach 2:
The system applies classification only to the extent necessary for each object size category using appropriately scaled models. Smaller objects are processed by lighter classifiers while larger objects use more computationally intensive classifiers, avoiding unnecessary computational resources being wasted on all objects uniformly.
3Reliability
If feature extraction methods are applied to improve detection accuracy, then false positive reduction improves, but processing time increases
Solution Approach 1:
The feature extraction and classification process is segmented by object size categories. Different feature sets and extraction methods are applied to small, medium, and large objects respectively, allowing the system to use computationally efficient features for small objects while applying more comprehensive feature analysis only to larger objects that require it.
Solution Approach 2:
The system applies different levels of feature extraction complexity locally to different object size categories. Rather than uniformly extracting all possible features from all objects, the system tailors the feature extraction depth to the specific requirements of each object size, reducing overall processing time while maintaining detection reliability.
Data Source
AI summary
Disclosed are methods, devices, and computer program products for red-eye detection in an image. In one example embodiment, a method for detecting red-eye objects in an image includes several acts. First, a set of candidate red-eye objects identified in the image is received. Then, features are extracted from the candidate red-eye objects and, with a plurality of classifiers, a false red-eye object is eliminated from the set of candidate red-eye objects based on the extracted features. First and second ones of the plurality of classifiers are optimized for classifying objects in a first range of sizes using first and second ones of the extracted features, respectively. Furthermore, third and fourth ones of the plurality of classifiers are also optimized for classifying objects using the first and second ones of the extracted features, respectively, but for objects in a second range of sizes.