Related Experiment Videos
Convolutional face finder: a neural architecture for fast and robust face detection
Christophe Garcia1, Manolis Delakis
1France Telecom R&D, 4 rue du Clos Courtel, 35512, Cesson Sivigne Cedex, France. cgarcia@csd.uoc.gr
IEEE Transactions on Pattern Analysis and Machine Intelligence
|November 4, 2004
Summary
This study introduces a new convolutional neural network for robust face detection, even with rotated and varied facial patterns in complex images. The system achieves high accuracy with few false positives without complex preprocessing.
Area of Science:
- Computer Vision
- Machine Learning
- Artificial Intelligence
Background:
- Face detection is crucial for many applications.
- Existing methods struggle with highly variable face patterns and complex backgrounds.
- Robust and efficient face detection remains a challenge.
Purpose of the Study:
- To develop a novel face detection approach using convolutional neural architecture.
- To achieve robust detection of highly variable and rotated face patterns in real-world images.
- To demonstrate an efficient system without costly local preprocessing.
Main Methods:
- Utilized a convolutional neural architecture for face detection.
- Employed automatic synthesis of problem-specific feature extractors from training data.
- Implemented a pipeline of convolution and subsampling modules processing the raw input image.
- Avoided hand-made feature design and assumptions about facial areas.
Main Results:
- Achieved robust detection of face patterns with variations up to +/-20 degrees in-plane rotation and +/-60 degrees out-of-plane.
- Demonstrated a very high detection rate with a particularly low level of false positives on difficult test sets.
- Showcased efficiency without requiring multiple networks for challenging cases.
Conclusions:
- The proposed convolutional neural network approach offers an efficient and robust solution for face detection.
- The system effectively handles significant variations in face patterns and complex image conditions.
- Eliminates the need for costly local preprocessing, simplifying the face detection pipeline.