Classification of protein crystallization imagery

We investigate automatic classification of protein crystallization imagery, and evaluate the performance of several modern mathematical tools when applied to the problem. For feature extraction, we try a combination of geometric and texture features; for classification algorithms, the support vector machine (SVM) is compared with an automatic decision-tree classifier. Experimental results from 520 images are presented for the binary classification problem: separating successful trials from failed attempts. The best false positive and false negative rates are at 14.6% and 9.6% respectively, achieved by feeding both sets of features to the decision-tree classifier with boosting.