Abstract

The determination of blood types is a crucial step in transfusion and diagnosis in the health sector. In this paper, imaging techniques coupled with deep learning algorithms are employed to automatically recognize blood groups. Connective Embedded with Scale Invariant Feature Transform (SIFT), Directional BRIEF and Rotated BARF (ORB) and spatial correlation of fingerprints using Gabor filters are applied to identify the distinguishing features of blood group as well as fingerprints images. The extracted features are then classified by Convolutional Neural Networks (CNNs). Also, fingerprint features are supplemented with ridge frequency and spatial features to further improve the determination of blood group. The framework consists of contrast enhancement and denoising techniques for image quality improvement, making it robust to image quality fluctuations. To increase model effectiveness and generalizability, transfer learning with VGG, ResNet, and DenseNet as base models is employed. Tested against different datasets, the method shows a good level of accuracy, consistency, and rate of success in recognizing blood types and discriminatory marks. This novel technique is fully automated. It could revolutionize the process of blood group and transfusions by allowing fast and accurate management of blood transfusions and patients.