Gender Recognition Based on Gradual and Ensemble Learning from Multi-View Gait Energy Images and Poses.

Area of Science:

• Computer Vision
• Machine Learning
• Biometrics

Background:

• Image-based gender classification is crucial for applications like surveillance and marketing.
• Current methods using gait energy images (GEI) face limitations due to similar intra-class appearances and lack of clear posture representation at certain angles.

Purpose of the Study:

• To propose a novel gender classification framework that integrates GEI with characteristic walking poses.
• To enhance gender classification accuracy by leveraging multi-view and multi-pose information.

Main Methods:

• A cascade network architecture was developed, comprising a multi-stream feature extractor and an ensemble learning-based classifier.
• The network is trained to extract gait features from multiple input images, including GEI and characteristic poses.
• The framework processes images acquired from multiple views to learn gait features progressively.

Main Results:

• The proposed framework demonstrated superior performance compared to state-of-the-art methods that rely solely on GEI or pose information.
• The integration of multiple inputs (GEI and poses) significantly improved classification accuracy on benchmark datasets.
• The cascade network effectively learned discriminative gait features from diverse visual inputs.

Conclusions:

• Combining GEI with characteristic walking poses offers a more robust approach to image-based gender classification.
• The proposed multi-input cascade network framework represents a significant advancement in gait-based biometrics.
• Future work can explore further enhancements by incorporating additional gait dynamics and advanced deep learning architectures.