Comparison of histogram-based feature sets for medical image modality categorization

This work is concerned with the automatic indexing of medical images according to their medical modality for image retrieval purposes inside the CISMeF health-catalogue. The paper investigates the extraction of an accurate modality signature from gray-level medical images based on various histogram weighting-schemes. The medical image database contains six main modalities and was selected by a medical specialist, from a real healthcare environment. The authors extracted and compared the relative contribution of different weighted histogram feature vectors in describing the visual content of medical images. The highest modality classification accuracy (78.67%) was obtained with the LBP (local binary pattern) weighted histogram, using a SVM classifier.