18.03.2014 Views

Recognition of facial expressions - Knowledge Based Systems ...

Recognition of facial expressions - Knowledge Based Systems ...

Recognition of facial expressions - Knowledge Based Systems ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

almost closed it obviously does not work. It creates a candidate list for the position <strong>of</strong> the<br />

eyes and further detects the most appropriate two pixels to the known eye area<br />

requirements. This can be avoided by setting a condition that finally detected pixels must<br />

have the intensity above a given value. The other way would be to create a structure for<br />

the alternative previous neural network to encode also the intensity <strong>of</strong> the candidate pixel<br />

on the input neural layer.<br />

Another limitation is that some calibration has to be done before the actual working<br />

session. It consists in adjusting all the parameters related to the searched areas and<br />

pixel/area characteristics.<br />

Figure 26. Model 's characteristic points<br />

So far only the positions <strong>of</strong> the pupils are detected. For the detection <strong>of</strong> the characteristic<br />

points <strong>of</strong> each eye (Figure 26/ P1, P3, P5, P7, P2, P4, P6, P8), some processing has to be<br />

done. A new parameter is computed as the distance between the pupils. That is used for<br />

scaling all the distances so as to make the detection person independent.<br />

The analysis starts again from the Sobel edge detector point. It further includes a<br />

threshold pixel removing. Given the position <strong>of</strong> the pupils for both eyes, the procedure is<br />

constrained to search for the pixels only in a given area around the pupil. Within the<br />

searched area, all pixels are analyzed through the procedure <strong>of</strong> computing the mean and<br />

variance as previously described. <strong>Based</strong> on the two values, some pixels are removed from<br />

the image.<br />

- 85 -

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!