3 Face recognition
3.2 Feature extraction

Some facial recognition algorithms are based on features extracted from an image of the subject's face – on facial features. For example, an algorithm may analyze the relative position, size, and/or shape of the eyes, nose, mouth, cheekbones, and jaw. These features are then used during search in group of images for matching features. Other algorithms normalize a gallery of face images and then compress the face data, saving only the data in the image that is useful for face recognition. A tested image is then compared with the face data.

Before feature extraction all images should be pre-processed and normalized.

As part of pre-processing is dimension reduction of all input images to defined size. Also contrast limited adaptive histogram equalization can be applied (CLAHE). The normalized images can be masked to omit the background and leave only the face region.

The main objective of the normalization process is to minimize the uncontrolled variations that occur during the acquisition process and to maintain the variations observed in facial feature differences between individuals.

What can bring differences into images is also the pose change.

Feature extraction involves reducing the amount of resources required to describe a large set of data. During face recognition, analysis of big amount of data is performed. Analysis with a large number of variables generally requires a large amount of memory and computation power. Feature extraction is related to reduction of variables and data.

For facial feature extraction the edge detection methods are used most frequently. Very good results are achieved also by local binary patterns (LBP).

Edge detection is the name for a set of mathematical methods where main aim is to detect points in digital image where brightness changes sharply. These image points with crisply change of brightness are typically organized into a set of curved line segments named edges.

The most frequently used functions for edge detections are Sobel operator (called also Sobel filter), Prewitt operator or Gabor filters.

Extracts features from pre-processed faces can be done via LBP histograms as features. LBP histograms are considered as one of the best features for recognizing faces even when only a limited number of samples is available and can be easily computed in the real time [5] (Fig. 2.1).

image
Fig. 2.1 – Example of result for feature extraction