/Ear-Recognition-Pipeline

Biometric pipeline for Ear Recognition

Primary LanguageJupyter Notebook

Ear-Recognition-Pipeline

In this work, we implement a ear recognition pipeline. Our pipeline includes training a custom DeepLabV3 segmentation model to isolate ears in images, extracting both Local Binary Patterns (LBPs) and ResNet50 (pre-trained on ImageNet) fea- tures from the segmented ears, and implementing a matching stage to identify individuals based on extracted features. The performance of our system is evaluated at each stage of the pipeline, including the segmentation, recognition, and overall pipeline stage. For further detail please read Biometric Pipeline

Project structure

Notebook trains and evaluates a custom DeepLabV3 segmentation model, the model is fine-tuned for the task of ear segmentation.

Notebook extracts two types of feature vectors. We extract Local Binary Patterns (LBP) and ResNet50 (pretrained on ImageNet) features.

Notebook implements and evaluates the recognition pipeline.