06.03.2013 Views

Artificial Intelligence and Soft Computing: Behavioral ... - Arteimi.info

Artificial Intelligence and Soft Computing: Behavioral ... - Arteimi.info

Artificial Intelligence and Soft Computing: Behavioral ... - Arteimi.info

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.

2.2.3 Feature-based Approach<br />

for Pattern Recognition<br />

The main consideration of this approach is to extract a set of primitive features,<br />

describing the object <strong>and</strong> to compare it with similar features in the sensory<br />

patterns to be identified. For example, suppose we are interested to identify<br />

whether character ‘H’ is present in the following list of characters (fig. 2.2 (b)).<br />

A H F K L<br />

Fig. 2.2 (b): A list of characters including H.<br />

Now, first the elementary features of ‘H’ such as two parallel lines <strong>and</strong> one<br />

line intersecting the parallel lines roughly at half of their lengths are detected.<br />

These features together are searched in each of the characters in fig. 2.2 (b).<br />

Fortunately, the second character in the figure approximately contains similar<br />

features <strong>and</strong> consequently it is the matched pattern.<br />

For matching facial images by the feature-based approach, the features like<br />

the shape of eyes, the distance from the nose tip to the center of each eye, etc.<br />

are first identified from the reference image. These features are then matched<br />

with the corresponding features of the unknown set of images. The image with<br />

the best matched features is then identified. The detailed scheme for image<br />

matching by specialized feature descriptors such as fuzzy moments [5] will be<br />

presented in chapter 23.<br />

2.2.4 The Computational Approach<br />

Though there exist quite a large number of literature on the computational<br />

approach for pattern recognition, the main credit in this field goes to David<br />

Marr. Marr [19] pioneered a new concept of recognizing 3-dimensional objects.<br />

He stressed the need for determining the edges of an object <strong>and</strong> constructed a<br />

2 ½-D model that carries more <strong>info</strong>rmation than a 2-D but less than a 3-D image.<br />

An approximate guess about the 3-D object, thus, can be framed from its 2 ½-D<br />

images.<br />

Currently, computer scientists are in favor of a neural model of perception.<br />

According to them, an electrical analogue of the biological neural net can be<br />

trained to recognize 3-D objects from their feature space. A number of training

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

Saved successfully!

Ooh no, something went wrong!