Activity 18: Minimum Distance Classification

September 17, 2008 at 9:05 pm (Uncategorized)

Pattern recognition techniques has long been studied by scientists for their potential applications.  Human face recognition, coral reef analysis and gait analysis have been successful implentations of pattern recognition techniques.

One of the most basic forms of pattern recognition techniques is through attibutes of a given sample and compare it with the attributes of a training set. In this activity, we use different samples of chips, namely Chippy, Nova and Piatos and attempted to distinguish chips from each other through the attributes obtained form the training set.

Ten samples of each kind of chip was photographed with a camera. After which, cut each images to ascertain that a single image contains a single chip. Image processing techniques such segmentation was performed to remove the chip from the background. Afterwich the average value of the R, G and B of each chip was obtained as well as the area. The chip is then represented by the vector x containing the features of the chip. We do this for all chips in our training set. If we have W types of chips (chippy, nova, and piatos), we obtain the average feature vector of each class given by

for all j=1,2,..W.

The features of each test chip are obtained and the distance from each class is computed using

where

The test chip is classified based on the class with the minimum distance. Take note also that I normalized each vector to eliminate bias towards features whose values are extremely large (area for example) and eliminate bias against small numbers (R value for example).

Here is the confusion matrix of the test set of my 3-class classification. This is good enough with accuracy of 87%.