SCIMD-6: Source Camera Identification — Mobile Devices Dataset

# 📷 SCIMD-6: Source Camera Identification — Mobile Devices Dataset

## 📂 Overview

**SCIMD-6** is a carefully curated image dataset developed at **Bapatla Engineering College** to support research in **source camera identification** using images from **mobile devices**. The dataset contains **6315 RGB images**, acquired from **six different smartphones** under **diverse real-world conditions**.

## 📱 Devices Used

Mobile Device

Number of Images

Moto G64 5G         

1006

Moto G85 5G         

1037

Nothing A001

1036

Realme 8 Pro        

1001

Redmi 14C 5G

1014

Xiaomi M2101K6P     

1221

Total

6315

📌 *Note*: Slight imbalance exists across classes but overall distribution is fairly uniform.

## 🌄 Image Characteristics

- 📐 **Resolution**: All images are resized to **224×224** pixels for compatibility with CNN architectures.

- 🌤️ **Conditions**: Captured in a variety of **uncontrolled environments**, including:

  - Indoor and outdoor

  - Sunny and rainy weather

  - Casual perspectives and variable lighting

- 🤳 **Capture Style**: Intentional lack of discipline in framing adds **real-world complexity** for model robustness testing.

## 📑 Included Files

- 📁  A zipped file consisting of `Motog64_5G/`, `Motog85_5G/`, ..., `Xiaomi_M2101K6P/`: Folders containing 224×224 RGB images per mobile device.

- 📄 `merged_common.csv`: A metadata file containing **EXIF information**  (Exchangeable Image File Format ) extracted from all images (e.g., Make, Model, ExposureTime, FocalLength).

## 🎯 Intended Use

This dataset is intended for tasks such as:

- 📸 **Source Camera Identification (SCI)**

- 🔬 **Image Forensics and Provenance Analysis**

- 🤖 **Fine-grained Classification and Transfer Learning**

- 🧠 **Deep Learning Model Benchmarking in Forensic Settings**

## 🧪 Benchmark Baseline

We provide a baseline experiment using **ResNet-50**, achieving an initial test accuracy of **80%** on this dataset. This suggests the dataset's **challenging and discriminative nature** despite class similarity.

📚 Potential Applications of the Dataset

This dataset, although primarily designed for source camera identification using mobile device images, supports a wide range of research directions and practical applications:

1. Source Camera Identification (SCI)

Classification of images based on the originating mobile device using intrinsic sensor characteristics.

Enables research in PRNU-based techniques and camera model/device fingerprinting.

2. Image Forensics and Metadata Consistency Analysis

Verification of metadata integrity using image content.

Detection of inconsistencies in EXIF fields such as shutter speed, ISO, focal length, and timestamp.

Applicable in detecting tampered or manipulated media.

3. Shutter Speed and ISO Estimation (Regression Tasks)

Pixel-to-metadata learning: predicting EXIF fields like ISO speed rating or exposure time directly from the image content.

Useful for modeling camera behavior and building metadata synthesis pipelines.

4. Image Quality Assessment (IQA) and Denoising

Training and benchmarking denoising models under real-world noise conditions (e.g., high ISO settings).

Correlation of EXIF parameters with perceptual quality for no-reference IQA research.

5. Environmental and Scene Classification

Scene-type inference (indoor/outdoor, sunny/cloudy, low-light conditions) based on visual content and EXIF cues.

Aids in tasks like environmental awareness, adaptive imaging, or low-light enhancement.

6. Image Provenance and Authorship Verification

Attribution of images to devices for media forensics and misinformation detection.

Combines device classification with temporal and spatial metadata for provenance tracing.

7. Training and Evaluation of Robust Vision Models

Offers real-world diversity in lighting, context, and device pipeline characteristics.

Supports robustness evaluation of CNNs, Vision Transformers, and vision-language models in uncontrolled environments.

The SCIMD-6 dataset is publicly available on multiple trusted platforms for broad accessibility and reproducibility:

## 📌 Citation

If you use this dataset in your research, please cite as:

@dataset{chandramohan2025scimd6,

  author       = {B. Chandra Mohan and Ch. Pavan Kumar and K. Sri Harsha and Ch. Nagaraju and Sandhyana T and Suvarna Lakshmi M},

  title        = {SCIMD-6:  Source Camera Identification Mobile Devices Dataset},

  year         = {2025},

  publisher    = {Zenodo},

  url          = {https://your-dataset-link-here},

  note         = {A benchmark dataset for source mobile camera identification with diversified conditions and EXIF metadata.}

}

---

## 📬 Contact

For inquiries or academic collaborations:

**Dr. Chandra Mohan Bhuma** 

Department of Electronics & Communication Engineering 

Bapatla Engineering College 

✉️ [e-mail]

## 🔒 License

This dataset is released under the **Creative Commons Attribution 4.0 International (CC BY 4.0)** license.