DT-Binarize: A Hybrid Binarization Method using Decision Tree for Protein Crystallization Images

A single thresholding technique may not provide the best binarization for all images of datasets such as protein crystallization images. To overcome this limitation, multiple thresholding methods are used to binarize images. Whenever multiple thresholding techniques are used, it is important to know which one provides the best result automatically. To solve this problem, in this study, we propose an alternative technique for image thresholding that employs a tree based structure to determine the best thresholding approach for a particular case. The leaf nodes of the tree indicate different global thresholding techniques, which have different abilities to binarize the image. We try to select the best approach by making decisions that are based on the characteristic features of the sample such as standard deviation. We have applied this technique to our protein image dataset and compared the results with the ground truth binary images that are manually generated by experts. Experimental results indicate that using a selecting the best one in a group of global thresholding methods is beneficial rather than single one. We provide the comparison results using some well-known accuracy measures. Our technique has reached 0.82 using Matthew’s correlation coefficient (MCC) and increased the MCC value by 0.11.