NusantaraVision - Image Processor & Traditional Weapon Detector
I completed this as a group final assignment for the Digital Image Processing course. The pipeline covers three stages: preprocessing a dataset of Indonesian traditional weapon images, training a classifier with HOG feature extraction and Euclidean Distance matching, then serving it through a Flask web interface for browser-based inference.
Main Features
- Image Preprocessing: Adjusts brightness, saturation, sharpness, and noise reduction via slider controls before an image enters the detection pipeline.
- Histogram Equalization: Normalizes pixel intensity distribution using equalization to improve input image contrast.
- Traditional Weapon Detection: Extracts texture features with HOG (Histogram of Oriented Gradients), then classifies the weapon type (Keris, Kujang, etc.) against a reference dataset using Euclidean Distance.
- Inference Result Visualization: Displays the enhanced image, HOG feature representation, identified weapon type, and Euclidean Distance rankings on the same page.
- Web-based Inference: The Flask interface accepts image uploads from users and returns preprocessing and classification results without running Python scripts manually.
Tech Stack
- Backend: Python, Flask
- Computer Vision: OpenCV
- Feature Extraction: HOG (Histogram of Oriented Gradients)
- Classification: Euclidean Distance
- Frontend: HTML, CSS, JavaScript
Project Links
- GitHub Repository: Isann22/PCD_Kelompok8
- Video Demo: Google Drive Demo