Knuckle classification has witnessed enormous progression in the last decade. The use cases of knuckle features to identify perpetrators of serious crime and for human authentication in consumer applications have paved new directions of research. The knuckle creases formed on the distal interphalangeal (DIP), proximal interphalangeal (PIP) and meta-carpo phalangeal (MCP) joints of hands possess unique properties. The capability of popular deep learning models to extract the discerning features of knuckle regions needs more experimentation. Hence, this research emphasizes on how different deep learning models perform in identifying the finger knuckle patterns and textures using an extensive dataset comprising of the base (or MCP), major (or PIP) and minor (or DIP) knuckles of all four fingers explicitly. We evaluated the results on the ResNet50, ResNet101, and DenseNet201 models. To calculate the loss metrics and accuracy of each model, it was first trained and then put to test. The outcomes of our evaluation show that these models achieve significant results on various kinds of knuckles and fingers. The best test case was found to be the major knuckle of the middle finger when trained with the DenseNet201 model with 75.38% accuracy.