Camera Sensor Noise Characterization

A Python toolkit for analyzing noise characteristics and dynamic range of digital camera sensors from raw image files.

Overview

This project provides tools to measure and compare the noise performance of different camera sensors by analyzing raw image files (DNG, ERF, etc.). It calculates the sensor's dynamic range in Exposure Value (EV) units and provides professional visualizations for comparison.

Features

• Automated Raw File Scanning: Process directories of raw images from multiple cameras
• Noise Analysis: Calculate standard deviation, mean, min, and max values from raw sensor data
• Dynamic Range Calculation: Compute Exposure Value (EV) as a measure of sensor dynamic range
• GPU Acceleration: Optional CUDA support for faster processing of large raw files
• Result Caching: Automatically saves and loads analysis results to avoid reprocessing
• Progress Tracking: Visual progress bars during file processing
• Professional Visualizations: Interactive plots with grouped legends and consistent styling
• Multi-Camera Comparison: Aggregate analysis across multiple camera models and variants

Installation

Prerequisites

• Python 3.8+
• ExifTool command-line tool
• NVIDIA GPU with CUDA (optional, for GPU acceleration)

Install ExifTool

Linux (Ubuntu/Debian):

sudo apt update && sudo apt install libimage-exiftool-perl

macOS:

brew install exiftool

Install Python Dependencies

pip install -r requirements.txt

Optional: GPU Acceleration

If you have an NVIDIA GPU with CUDA installed:

# Check your CUDA version
nvcc --version

# Install CuPy (for CUDA 12.x/13.x)
pip install cupy-cuda12x

Quick Start

Basic Usage

import sensor_camera
from collections import OrderedDict

# Create Analysis object with path to your raw files
analysis = sensor_camera.Analysis('path/to/raw/files')

# Define cameras to scan
scan_specs = OrderedDict([
    ('Leica M11 (60MP)', {'path': 'M11-60MP', 'suffix': 'DNG'}),
    ('Leica Q3 (36MP)', {'path': 'Q3-36MP', 'suffix': 'DNG'}),
])

# Scan all cameras
scan_results = analysis.scan(scan_specs)

# Create aggregate dataset
data = analysis.create_aggregate()

# Save results
analysis.save_aggregate('results.csv')

# Create plots
figures = analysis.plot_ev_vs_iso()  # EV vs ISO for all exposure times
for fig in figures:
    fig.show()

API Documentation

Sensor Class

The Sensor class handles scanning individual camera directories:

sensor = Sensor(path='path/to/camera/files')
data = sensor.scan(path='subdirectory', suffix='DNG', force_rescan=False)

Parameters:

• path: Base directory or subdirectory containing raw files
• suffix: File extension to scan for (e.g., 'DNG', 'ERF')
• force_rescan: If True, rescan even if cached results exist

Features:

• Automatically crops sensor data for known camera artifacts
• Extracts EXIF metadata (ISO, exposure time, dimensions)
• Computes noise statistics (std, mean, min, max)
• Calculates Exposure Value (EV)
• Saves results to noise_results.csv in the scanned directory

Analysis Class

The Analysis class manages aggregate analysis across multiple cameras:

analysis = Analysis(base_path='path/to/data')

Methods

scan(scan_specs, force_rescan=False)

• Scans multiple cameras according to specifications
• Returns OrderedDict with results for each camera

create_aggregate(camera_list=None)

• Combines data from selected cameras into single DataFrame
• Preserves camera variant information

save_aggregate(filename='aggregate_analysis.csv')

• Saves aggregate data to CSV file

plot_ev_vs_iso(data=None, exposure_time=None, title=None, height=700, ev_range=None)

• Creates professional EV vs ISO plot(s)
• exposure_time: Single value, list, or None (all times)
• Returns single figure or list of figures

plot_ev_vs_time(data=None, iso=None, title=None, height=700, ev_range=None)

• Creates professional EV vs Exposure Time plot(s)
• iso: Single value, list, or None (all ISOs)
• Returns single figure or list of figures

get_aliases(short_names=None)

• Creates convenient variable aliases for backward compatibility

Exposure Value (EV) Calculation

The sensor's dynamic range is quantified as:

$$\text{EV} = \log_2\left(\frac{W - B}{\sigma}\right)$$

Where:

• $W$ = white level (sensor saturation point)
• $B$ = black level (sensor baseline)
• $\sigma$ = standard deviation of raw sensor data (noise)

This represents the signal-to-noise ratio in stops (each stop = 2× light).

Supported Cameras

The toolkit includes optimized handling for:

• Leica: M11, M10, Q, Q3, SL2-S, CL, TL2, M Monochrome
• Ricoh: GR III, GXR
• Epson: R-D1
• Apple: iPhone 13 Pro
• Pixii: Pixii cameras

Camera-specific crops are applied to remove known sensor artifacts.

Visualization Features

Professional Plots

• Grouped Legends: Camera variants grouped by base model
• Consistent Colors: Same color for all variants of a model
• Line Style Differentiation: Different dash patterns for variants
• Interactive: Hover to compare, click to toggle cameras
• Auto-scaling: Y-axis automatically adjusts to show all data
• Publication Ready: Clean styling suitable for presentations and papers

Plot Types

1. EV vs ISO: Shows how dynamic range degrades with higher ISO
2. EV vs Exposure Time: Shows noise characteristics at different shutter speeds

Project Structure

camera-char/
├── sensor_camera.py       # Main module with Sensor and Analysis classes
├── sensor_noise.ipynb     # Jupyter notebook for analysis
├── requirements.txt       # Python dependencies
├── aggregate_analysis.csv # Combined results from all cameras
└── detector-noise/        # Directory containing raw image files
    ├── M11-60MP/
    │   ├── *.DNG
    │   └── noise_results.csv
    ├── Q3-36MP/
    └── ...

Performance

GPU Acceleration

With CuPy installed, statistical calculations are performed on GPU:

• 5-10x speedup for large raw files (20-60MP)
• Automatic fallback to CPU if GPU unavailable
• Explicit memory management to prevent GPU memory issues

Caching

Results are automatically cached in noise_results.csv:

• First scan: Processes all raw files (slower)
• Subsequent scans: Loads from CSV (instant)
• Use force_rescan=True to regenerate

Contributing

This is a research project for camera sensor noise analysis. Contributions are welcome!

License

See repository for license information.

References

• ISO 12232 - Photography — Digital still cameras — Determination of exposure index, ISO speed ratings, standard output sensitivity, and recommended exposure index

Author

Created for systematic camera sensor noise characterization and comparison.