Performance comparison of ORB, SURF and SIFT using Intracranial Haemorrhage CTScan Brain images

Abstract

Medical images are crucial for both the doctor's accurate diagnosis and the patient's subsequent therapy. It is feasible to swiftly identify lesions in medical photos by using clever algorithms, and it is crucial to extract information from images. Feature extraction is an important step in image classification. It allows the representation of the content of images as perfectly as possible. The intention of this study is to certain overall performance assessment among the feature detector and the descriptor method, especially while there are numerous combos for assessment. Three techniques were decided on for the feature descriptors: ORB (Oriented FAST and Rotated BRIEF), SIFT (Scale Invariant Feature Transformation), and SURF (Accelerated Robust Feature) and to calculate matching evaluation parameters, for example, the number of key points in the image, Execution time required for each algorithm and to find the best match. The dataset was taken from Kaggle, which contained 170 CTScan images of the brain with intracranial hemorrhage masks. The brute force method is used to achieve feature matching. Performance analysis shows the discriminative power of various combinations of detector and descriptor methods. SURF algorithm is the best and most robust in CTScan imaging to help medical diagnosis.