Malaria Cell Detection using HOG

Project Summary

Classical computer vision pipeline to detect malaria-infected cells using HOG (Histogram of Oriented Gradients) features and an SVM classifier. Includes image preprocessing, feature extraction, model training with cross-validation, evaluation, and a simple Streamlit UI for inference.

Skills Demonstrated

Image Preprocessing HOG Feature Extraction SVM Classification Model Selection Cross‑Validation Hyperparameter Tuning Evaluation (Accuracy/F1) Model Persistence Streamlit UI

Tools Used

Python 3.8+ OpenCV scikit-image (HOG) Scikit‑Learn (SVM) NumPy Matplotlib Joblib Streamlit Hugging Face Spaces

Solution

A lightweight CV pipeline: preprocess images, extract HOG descriptors, train an SVM with cross‑validation and grid search, evaluate on a held‑out set, persist the model (joblib), and serve predictions with a Streamlit app.

Approach

Data Prep: Normalize/resize images; optional denoising and color to gray.
Features: Compute HOG descriptors with tuned cell/block sizes.
Modeling: Train SVM (RBF/linear) with stratified CV.
Tuning: Grid search C, gamma, HOG params.
Evaluate: Accuracy/F1 and confusion matrix.
Persist & Serve: Save model and run Streamlit UI.

Project Link(s)

Repository: Malaria Cell Detection (HOG)

Streamlit App (Hugging Face Space)

Completed: 2025