Activity 15: Probabilistic Classification

Objectives: To apply Linear Discriminant Analysis to classify and sort objects from two distinct classes.

Tools: Scilab and data from Activity 14.

Procedure: The aim of the activity is to classify the objects in an

image into distinct groups in much the same way as the previous activity. Instead of using the distance method, we use linear discriminant analysis or LDA, the details of which can be found elsewhere. We utilize the same set of images from the previous activity, that of the coins.

As mentioned the data from the previous activity is used with same classification characteristics, that is area and color. Since we want to examine the usage of LDA on two classes of objects only, we sort our data set according to area and exclude the first ten values, as these correspond to one class of objects. This is to eliminate the need to mask one of the object test classes and redoing the image processing routines.

We then apply the LDA routine per the tutorial provided and plot in two axes, we get a figure similar to the previous activity. The crosses represent the loci of the objects in the image with respect to area and color.

Note the distinctive separation of the two classes with the lower set representing the lower denomination coins. Also of note is the better discrimination between the two classes compared to the previous activity. In the said procedure, the distinctions between two almost similar classes is unclear. In this case the deliniation is much clearer and hence it can be expected that this method will yield a better grouping and classification result.

Evaluation: For the proper classification of the objects a grade of 10 is appropriate.

Acknowledgements: The usual thanks to Neil Cabello and Earl Panganiban for the unfailing assistance.