Training face detector using adaboost algorithm
Face detection is a complex and challenging task due to the high variability in faces and amongst faces. Also for a given image, a face detector should be able to identify and locate all faces, regardless of their position, scale, orientation (up-right, rotated) and pose (frontal, profile). To reduc...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
2009
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/17909 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Face detection is a complex and challenging task due to the high variability in faces and amongst faces. Also for a given image, a face detector should be able to identify and locate all faces, regardless of their position, scale, orientation (up-right, rotated) and pose (frontal, profile). To reduce the complexity associated with this field, this project will focus on detecting upright frontal faces.
In this project, the method of face detection developed by Paul Viola and Michael Jones was chosen to be explored in and based on. A large dataset of training images was used during the process of training the face detector. An exhaustive number of Haar-like features were extracted from these training images but using an AdaBoost algorithm, only a few were selected as useful to the face detector in differentiating the faces from the non-faces. Being trained by these useful features, the face detector’s performance was evaluated and analysed on testing images under varying parameters to determine the best conditions for optimal performance. Modifying these parameters based on these results, the face detector was then implemented on real images and its performances evaluated. The process of doing this project has provided a deeper understanding for the face detection process. |
|---|