How shivammitra/fingerprint-based-blood-group-detection works
This repository is a research/prototype system for predicting blood group from fingerprint imagery, with supporting model experiments and interpretability artifacts.
At a high level, the repo is organized around four subsystems:
1. Data package
• Dataset/ contains eight compressed datasets, one per blood group class:
- A+.zip, A-.zip, AB+.zip, AB-.zip, B+.zip, B-.zip, O+.zip, O-.zip
• This strongly suggests the core supervised-learning problem is an 8-class classification task.
• The zipped format indicates the raw images are likely stored externally from the code and unpacked for training or evaluation.
2. Model experimentation and training
• fingerprint_detection/ contains Jupyter notebooks for multiple CNN architectures:
- Alexnet.ipynb
- Lenet.ipynb
- Resnet.ipynb
- Vgg16.ipynb
• These notebooks are likely where model-specific preprocessing, training, validation, and comparison happen.
• The architecture boundary here is “experiment notebooks per backbone,” rather than a single consolidated training service.
• The naming implies the repo compares several deep learning feature extractors on the same fingerprint dataset.
3. Prediction orchestration
• ensemble_prediction.py is likely the integration layer that combines outputs from the individual model experiments.
• This file probably serves as the main inference or aggregation script, taking predictions from multiple networks and producing a final blood-group classification.
• In the architecture, this sits above the single-model notebooks and represents the ensemble decision boundary.
4. Explainability and domain analysis
• xai.py suggests explainable AI tooling for model interpretation.
• ridge_analysis/ contains one markdown document per blood group:
- blood_group_A+.md, etc.
• This indicates the repo includes post hoc analysis focused on ridge features or fingerprint ridge patterns, likely used to connect model behavior with fingerprint characteristics.
• These artifacts are downstream of the trained models and are intended for interpretation/reporting rather than training.
Likely end-to-end data flow
• Fingerprint images are organized by blood-group label in Dataset/.
• Individual CNNs are trained and evaluated in the notebooks under fingerprint_detection/.
• Their outputs are combined in ensemble_prediction.py to produce the final class prediction.
• xai.py generates explanation outputs to show why a prediction was made.
• ridge_analysis/*.md captures class-specific qualitative analysis of ridge patterns or findings.
Repo boundaries and architecture style
• This is not a multi-service application; it is a machine-learning research repository.
• The main boundary is between:
- raw data,
- model training experiments,
- prediction aggregation,
- interpretability/reporting.
• The code appears to be organized for experimentation rather than production deployment.
• There is no evidence of a database, message broker, web backend, or external API integration in the file tree.
Relevant technologies implied by the structure
• Python for orchestration and analysis scripts.
• Jupyter notebooks for model development and experimentation.
• Deep learning CNN architectures for image classification.
• Markdown for analysis documentation.
• ZIP archives as the dataset distribution format.
If you are drawing an architecture diagram, the most accurate depiction would show:
• dataset archives as the input layer,
• multiple parallel model-training branches for AlexNet/LeNet/ResNet/VGG16,
• an ensemble aggregation layer,
• a separate explainability/analysis layer,
• and markdown-based ridge-analysis outputs as the final interpretive artifacts.
Main entry points: ensemble_prediction.py Ensemble · xai.py XAI · Dataset Blood group data · Alexnet.ipynb AlexNet
Components 9
Data
Compressed labeled fingerprint datasets for the 8 blood-group classes.
- Blood group data · Eight ZIP archives, one per blood-group class, used as the supervised-learning input.
Dataset
Model Experiments
Notebook-based CNN training and evaluation branches.
- AlexNet · Experiment notebook for AlexNet-based fingerprint classification.
fingerprint_detection/Alexnet.ipynb - LeNet · Experiment notebook for LeNet-based fingerprint classification.
fingerprint_detection/Lenet.ipynb - ResNet · Experiment notebook for ResNet-based fingerprint classification.
fingerprint_detection/Resnet.ipynb - VGG16 · Experiment notebook for VGG16-based fingerprint classification.
fingerprint_detection/Vgg16.ipynb
Prediction
Aggregation and application-facing prediction flow.
Explainability
Post hoc interpretation and ridge-pattern analysis artifacts.
- XAI · Explainability tooling for interpreting model predictions.
xai.py - Ridge analysis · Class-specific ridge-analysis notes for all eight blood groups.
ridge_analysis
Connections 16
- Blood group data → AlexNet: train/eval data · Fingerprint images feed the AlexNet experiment.
- Blood group data → LeNet: train/eval data · Fingerprint images feed the LeNet experiment.
- Blood group data → ResNet: train/eval data · Fingerprint images feed the ResNet experiment.
- Blood group data → VGG16: train/eval data · Fingerprint images feed the VGG16 experiment.
- AlexNet → Ensemble: model output · AlexNet predictions are one input to the ensemble.
- LeNet → Ensemble: model output · LeNet predictions are one input to the ensemble.
- ResNet → Ensemble: model output · ResNet predictions are one input to the ensemble.
- VGG16 → Ensemble: model output · VGG16 predictions are one input to the ensemble.
- Ensemble → App entry: final prediction · The ensemble is surfaced through the application entrypoint.
- Ensemble → XAI: explain · Prediction results are interpreted with XAI tooling.
- AlexNet → XAI: inspect · Individual model behavior may also be analyzed.
- LeNet → XAI: inspect · Individual model behavior may also be analyzed.
- ResNet → XAI: inspect · Individual model behavior may also be analyzed.
- VGG16 → XAI: inspect · Individual model behavior may also be analyzed.
- XAI → Ridge analysis: report · Explainability findings are captured in ridge-analysis notes.
- Blood group data → Ridge analysis: class context · The class-labeled dataset provides the domain context for the reports.