Face Detection Ensemble with Methods Using Depth Information to Filter False Positives.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Face detection is a critical challenge in computer vision.
• Existing methods often struggle with high false positive rates in unconstrained environments.

Purpose of the Study:

• To develop an ensemble face detection system that maximizes true positives and minimizes false positives.
• To introduce novel filtering techniques for enhancing detection accuracy.

Main Methods:

• An ensemble of six distinct face detectors was created.
• Filtering steps utilizing depth map characteristics and wavelet processing were applied to candidate face subwindows.
• The ensemble's performance was evaluated on a combined dataset and the BioID benchmark.

Main Results:

• The proposed filtering steps effectively reduced false positives without compromising the face detection rate.
• The best ensemble achieved a 100% detection rate on the BioID dataset with minimal false positives.

Conclusions:

• The developed ensemble face detection system demonstrates superior performance in reducing false positives.
• The integration of depth map and wavelet-based filtering offers a robust approach for accurate face detection in diverse conditions.