Real Time Human Face Detection

Hits: 0
Research Area: Uncategorized
Status: Not started  
Funded by: Fundação para a Ciência e a Tecnologia Funding: 0.00 EUR

Real Time Human Face Detection

This work was developed under the project “BioRec: Non-Cooperative Multimodal Biometric Recognition”. The goal was to develop a system for human face detection in images, capable of real time response. There are several approaches to the problem, either by using images in gray scale, color, through edge detection or face geometry.

Used methods:
The main used algorithm was the one developed by Viola and Jones [1], based on gray scale images. This choice was supported by:

  • it’s speed and low error rates;
  • the use of simple features, that are invariant to scale;
  • the introduction of a new concept: integral image.

The training algorithm to select/combine simple features is a variant of AdaBoost algorithm.

Image 1: Features presented by Viola and Jones [1].
Image 2: Representation of integral image. The value of the (x,y) pixel is given by the sum of pixels above and on the left of it.

The changes made to the Viola and Jones method [1] were the extraction of new features, based on triangles. This allowed the presentation of 4 new types f features, represented in image 3.
For the evaluation of these new features in real time, 4 new types of integral images were created, found in image 4.

Image 3: Triangle based features.
Image 4: Representation of the integral triangular image. The value of (x,y) pixel is given by the sum of values above the main diagonal and the left point.

To this project, a series of training/test images were obtained from the internet. Those images were cropped and resized to the size of 24x24 pixels, making a set of 30 000 images total (divided into 10 000 faces and 20 000 no faces).

Detection process:

The detection process is presented in image 5.

Image 5: Detection process presented by Viola and Jones [1].


Table 1: Minimal EER’s values for the test set.

Image 6: Detector test result.

[1] Paul Viola and Michael J. Jones. Robust Real Time Face Detection. International Journal of Computer Vision, 57(2):137-154,2004.