In this paper, we describe our method for classification of protein crystallization trial images using geometric features. The objective is to automatically categorize a protein crystal according to the presence of protein crystal types in the images. We consider only the images consisting of protein crystals for the classification. The images are classified into 4 categories- needles, small crystals, large crystals and other crystals. Image classification consists of two main steps - image feature extraction and applying decision tree classifier. Our feature extraction includes application of canny edge detection, extraction of edge related features from the edge image, and extraction of blob related features from multiple thresholding techniques. We performed our experiments on 212 expert labeled images and tested our results using 10-fold cross validation. Our results indicate that the proposed classification technique produces a reasonable classification performance. The overall accuracy of the classification is 75%.
展开▼
