Camera Model Identification with a Data-Driven Model

Juan Calderón · Dana Villamizar · Sneider Sánchez

Digital Image Processing project at UIS 2025-1

• Visit the report for more information about the project.
• Visit the slides for the presentation.

Overview

This repository contains the source code and trained models for the project titled "Camera Model Identification with a Data-Driven Model". The goal of the project is to classify the smartphone brand or model that captured a digital image using convolutional neural networks.

The project implements two main strategies:

1. Replication of a CNN architecture proposed by Baroffio et al. for identifying traditional camera models.
2. Fine-tuning of a pre-trained EfficientNetV2-M model on a smartphone image dataset (FloreView) for brand-level classification.

The model achieves a validation accuracy of 82% on the FloreView dataset and is available through a simple web interface.

A detailed explanation of the methodology and results can be found in the final paper (PDF).

---

Datasets

• Training 1: Dresden Image Database
• Training 2: FloreView Dataset – smartphone images acquired under controlled conditions.

---

Model Options

The project includes two model implementations:

• CameraConvNet: Custom CNN from Baroffio et al., trained from scratch and later fine-tuned on smartphone images.
• EfficientNetV2_M: Pre-trained on ImageNet and fully fine-tuned on the FloreView dataset for mobile camera classification.

🔗 Download Trained Models
You can download the .pth files for both models from the following Google Drive folder:
Google Drive – Trained Models

---

Installation

To set up the environment and install dependencies:

git clone https://github.com/JJCG25/An-Image-Processing-Project-Camera-Model-Identification-with-a-Data-Driven-Model
cd An-Image-Processing-Project-Camera-Model-Identification-with-a-Data-Driven-Model
pip install -r requirements.txt