BINARY DESCRIPTOR BASED MULTI-CLASS OBJECT CATEGORIZATION FOR THYROID NODULE IMAGE ANALYSIS

In this research paper, an improved method to identify the object class in thyroid nodule visual images using multi-class classification is presented. A general method for object categorization is to quantize feature descriptors into visual words and then form a bag-of-feature model. Bag of Visual Word (BoVW) based Image Retrieval is a well-received method for Content Based Image Retrieval (CBIR). After this, multi-class classification is used to categorize objects into different classes. Feature detection and description act as the backbone of multi-class classification. BRISK & FREAK (Binary Robust Invariant Scalable Key points and Fast Retina Key point) are well-known binary descriptors in many computer vision applications. in this research paper proposed to use the combination BRISK with BRISK (BRISK-BRISK) and BRISK with FREAK (BRISK-FREAK) as binary detector and descriptor under the framework of SVM (Support Vector Machine) classification. The classification of thyroid nodule image visual categories using the proposed algorithm and their performance is evaluated based on the confusion matrix. A comparison of these methods with different kernels of the SVM classifier is also observed.