Leveraging Graphics Hardware for an Automatic Classification of Bone Tissue

Abstract

Zernike moments are fundamental digital image descriptors used in many application areas due to their good properties of orthogonality and rotation invariance, but their computation is too expensive and limits its application in practice, overall when real-time constraints are imposed. The contribution of this work is twofold: Accelerate the computation of Zernike moments using graphics processors (GPUs) and assess its expressiveness as descriptors of image tiles for its subsequent classification into bone and cartilage regions to quantify the degree of bone tissue regeneration from stem cells. The characterization of those image tiles is performed through a vector of features, whose optimal composition is extensively analyzed after testing 19 subsets of Zernike moments selected as the best potential candidates. Those candidates are later evaluated depending on its ability for a successful classification of image tiles via LDA, K-means and KNN classifiers, and a final rank of moments is provided according to its discriminative power to distinguish between bone and cartilage. Prior to that study, we introduce a novel approach to the high-performance computation of Zernike moments on GPUs. The proposed method is compared against three of the fastest implementations performed on CPUs over the last decade using recursive methods and the fastest direct method computed on a Pentium 4, with factor gains up to 125 ×on a 256 ×256 image when computing a single moment on a GPU, and up to 700 ×on a 1024 ×1024 image when computing all repetitions for a given order using direct methods.