Colour-Science face detection and tracking technology

(Where are faces in an image?)

The Colour-Science face detection engine is fully integrated in the i2e library. The library will analyse the images, enhance them and then export all data of the faces found inside an image.

The following data is available:

1. number of faces

2. pixel coordinates of every face rectangle

3. total detected face surface

We use a 22 stage haar classifier for the detection of faces. The detection is made only on the luminance image and a second validation cascade will make sure that there are not to much false positives.

Samples of face detection applications

	The Colour-Science i2e face-detection will detect also multiple faces inside one image. The detection will also work for +-90° rotated images. The algorithm gives you information about face coordinates, face size and face number. It is for example easy to use the face detection to sort your images by the number of detected people. The recognition rate for well visible faces is very high and quite fast. It is also possible to use this recognition in real time video applications.
	Density and color correction of portrait images can be much improved when a face is detected. The image enhancement algorithm knows now which the important object in the image is. The skin brightness and color inside the detection rectangle can so be tuned to an optimum skin tone.
	Many people rotate the camera for portrait images. The result is that such digital images are 90° rotated. Because the face detection algorithm recognizes the rotation it is simple to rotate those images automatically in upright position.
	Red eye reduction can be made much more securely if we know where the eyes are. Errors like “nose or lips removal” can be avoided. A very common error is also that only one of two red eyes is removed. All those errors can result in very costly customer complaints. This is why red eye removal is until now not automatically used in big labs. Face detection however can reduce the error rate dramatically so that red eye removal can be applied to all images.
	Samples of the Colour-Science face database. Several thousand faces are needed to train the object recognition engine.

Possible applications of face detection

· classification of images
It is very interesting to be able to sort an image archive by the number of detected people. Se also -> Colour-Science image classification technology

· image enhancement
for image enhancement it is very important to detect faces and therefore knowing the exact face color.

· automatic image rotation
Because the face detection algorithm recognizes the rotation it is simple to rotate those images automatically in upright position.

· red eye removal
Red eye reduction can be made much more securely if we know where the eyes are

· security
It is for example possible to detect people (faces) from a surveillance camera video stream and to generate automatic alarms.

· portrait / passport photography
Using the exact coordinates of the face rectangle it is possible to automatically crop portraits so that the faces are always in the same position.