The object detector described in [ Viola01] and [ Lein02] is based on Haar classifiers. Each classifier uses

k

rectangular areas (Haar features) to make decision if the region of the image looks like the predefined image or not. Figure

“Types of Haar Features”

shows different types of Haar features.

In the Intel IPP Haar features are represented using

IppRect

structure. Figure

“Representing Haar Features”

shows how it can be done for common and tilted features.

When the classifier

K

t

is applied to the pixel (

i

,

j

) of the image

A

, it yields the value

val1(t)

if

and

val2(t)

otherwise.

Here

w

l

is a feature weight,

norm(i, j)

is the norm factor (generally the standard deviation on the rectangle containing all features),

threshold(t)

,

val1(t)

and

val2(t)

are parameters of the classifier. For fast computation the integral representation of an image is used. Haar classifiers are organized in sequences called

stages

(

classification stages

). The stage value is the sum of its classifier values. During feature detecting stages are consequently applied to the region of the image until the stage value becomes less than the threshold value or all stages are passed.