Page 10 - FREE_TRIAL6Months

ARTIFICIAL INTELLIGENCE PART 3 PAGE 1/17
ROTATED FACIAL RECOGNITION BY BOIAN MITOV

In the previous article, I showed
you how you can use Radial Basis
Function classifiers to recognize
faces. In the example we provided
an image with the face, and the
classifier attempted to recognize
it. This was a nice simple example,
but for real applications we often
need to recognize faces or other
objects somewhere in an image,
and with different rotation.
To find the objects in a bigger
image, we will need to scan the
image with some X and Y step. If
the faces are also rotated as in
this image:

Then the task becomes even more complex. One of the options is to scan with some angle step rotation
of the image. The good news is that VisionLab includes the TVLSampler component specifically
designed to perform such X, Y and rotation image scanning, and IntelligenceLab is designed to work
well together with VisionLab.

All of the Mitov Software libraries are available fully functional, and free for non commercial and
evaluation purposes. You can download the IntelligenceLab and VisionLab and use them to test the
code from the article.

There are number of solutions to handle rotated images, but we will start with the simplest one –
using the TVLSampler to handle scanning with rotation, and use a simple non rotated training.
The IntelligenceLab classifiers can save training to file for later usage. This allows us to train once, and
use the training later.
First we will create a simple project to generate and save the training into a training file.

22 Issue Nr 8 2016 BLAISE PASCAL MAGAZINE