Spring semester update 10

[akzysk]

Improved "one button" user interface; it now builds both the training and test set according to the images the user checks for each set (DataSetBuilder.java). Working on building and evaluating a Weka classifier internally in Java (WekaTester.java), this will probably require some bug fixes with regard how the .arff files are saved, deleted, and managed.

Need to learn how to classify an image that is not "supervised", i.e. it is pre-classified by the user when uploaded. This will require some infrastructure; potentially adding another radio button on image upload to direct the program to classify the image.

Fully implemented the following Segment attributes: relativeX, relativeY, relativeArea, segmentCount. These attributes are more "logistical" in nature and are based on assumptions about sample images, and segments within the context of the full image, rather than the actual content of the individual segments.

A few test runs with ~20 images and 3500+ segments yield good results: 80%-90% accuracy on the segment level. Good attributes: convexity, circularity, stdDevR, relativeX, relativeY, relativeArea. Need to look into how test runs can be better tested and standardized. Can not yet find a huge dataset of bloodstains online.

Making my own bloodstain samples doesn't sound unreasonable, considering the samples of sweat and tears I've already contributed.