Adaboost face detector based on Joint Integral Histogram and Genetic Algorithms for feature extraction process.

Area of Science:

• Computer Science
• Machine Learning
• Artificial Intelligence

Background:

• Machine learning, particularly boosting techniques, is widely used for object detection due to low false positive rates.
• AdaBoost is popular for face detection but faces challenges in feature selection, leading to computational complexity and high memory usage.
• Conventional feature selection based on minimizing classification errors is inefficient for large feature sets.

Purpose of the Study:

• To propose a new method for training effective object detectors by discarding redundant weak classifiers.
• To enhance AdaBoost training by incorporating Genetic Algorithms (GA) for feature selection.
• To improve feature extraction using Joint Integral Histograms for more powerful features.

Main Methods:

• Modified AdaBoost training incorporating Genetic Algorithms (GA) to decouple feature selection from weak learner training error.
• Utilizing Joint Integral Histograms for enhanced feature extraction.
• Training and evaluating the proposed method on human face detection datasets.

Main Results:

• The proposed method requires a smaller number of weak classifiers compared to conventional algorithms.
• Achieved higher learning and faster classification rates.
• Outperformed state-of-the-art cascade methods in detection rate and false positive rate, with significant reduction in weak classifiers per stage.

Conclusions:

• The novel approach effectively trains object detectors by optimizing feature selection and extraction.
• The method offers a more efficient alternative to conventional AdaBoost for face detection and other object detection tasks.
• This research contributes to more efficient and accurate object detection systems through advanced machine learning techniques.