This is a code for ear based user identification assignment, which was a part of Image based biometry course at Faculty of Computer and Information Science, University of Ljubljana.
The assignment was implemented on AWE dataset [1] with perfectly cropped ear and also briefly tested on dataset that was obtained during the ear segmentation assignment [3]. I tested Pixel 2 pixel method, Local Binary Patterns (LBP) and two CNNs: ResNet50 and AlexNet. The starting point for my CNNs was [2]. I also used different image preprocessing techniques.
I obtained the following accuracies:
The rank curves on perfectly detected ears without preprocessing (except the required one for CNNs) are:
[1] http://ears.fri.uni-lj.si/datasets.html#awe-full
[2] https://pytorch.org/tutorials/beginner/finetuning_torchvision_models_tutorial.html