SCIMD-6: Source Camera Identification — Mobile Devices Dataset
Description
**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**. The mobile devices used and the number of images are Moto G64_ 5G(1006 images)Moto G85_5G(1037images), Nothing A001(1036 images),Realme 8 Pro(1001 images), Redmi 4C_5G(1014 images), Xiaomi M2101K6P(1221 images). Total: 6315 images. This dataset is also made available in other trusted public dataset repositories for better transparency and reproducibility.
Files
Steps to reproduce
📌 *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 zip file SCIMD-6.zip consisting of `Motog64_5G/`, `Motog85_5G/`, ..., `Xiaomi_M2101K6P/`: Folders containing 224×224 RGB images per mobile device is made available. - 📄 `merged_common.csv`: A metadata file containing **EXIF information** (Exchangeable Image File Format ) extracted from all images (e.g., Make, Model, Exposure Time, Focal Length). ## 🎯 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** 📚 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.