Sensor Noise Pattern Extraction for Imaging Device Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for identifying the imaging device that acquired an electronic image are unreliable and limited in their ability to distinguish between devices of the same make and model, especially when dealing with processed images.
Innovation Solution
The technique utilizes the unique stochastic pattern noise of digital imaging sensors, such as CCD, CMOS, and JFET, to identify the device by determining and correlating the sensor's reference noise pattern from images, allowing for reliable identification even from processed images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If header information or JPEG quantization matrices are inspected for device identification, then the identification process is simple, but the reliability is low because this information can be easily modified or lost during processing
Solution Approach 1:
The patent extracts and utilizes the sensor's inherent noise pattern as the identifying characteristic, separating this unique sensor signature from the image content itself. By focusing on the noise pattern rather than header information, the method achieves both simplicity in implementation and high reliability in identification, as the noise pattern cannot be modified or lost during standard image processing operations
2Reliability
If authentication watermarks are inserted into images for identification, then identification reliability is improved, but the method is only applicable to a very limited number of devices
Solution Approach 1:
The patent employs the sensor's own inherent noise pattern as the identifying characteristic, eliminating the need for external watermarks or authentication mechanisms. Each sensor naturally produces its unique noise pattern during image capture, making the method universally applicable to all digital imaging devices without requiring device-specific implementation or modification of the imaging process
3Reliability
If local pixel defects are used to identify a particular device, then device identification is achieved, but the method fails for cameras without pixel defects or that remove defects during post-processing
Solution Approach 1:
The patent shifts from using discrete, binary characteristics like pixel defects (present or absent) to using continuous, inherent parameters like noise patterns that all sensors possess. The noise pattern's statistical properties provide a reliable identifying characteristic that exists in all digital sensors regardless of their defect status or post-processing capabilities, greatly expanding the method's applicability
4Reliability
If dark current is used for camera identification, then identification is possible, but the reliability is limited because dark current is a relatively weak signal and can only be extracted from dark frames
Solution Approach 1:
The patent extracts the noise pattern from the actual image content itself, rather than requiring separate dark frames or specialized conditions. By utilizing the noise inherent in normally captured images, the method achieves both high reliability in identification and broad practical applicability across all standard imaging operations
Data Source
AI summary
A new technique for identifying whether images are derived from a common imager, e.g., a camera, or other imaging device such as a scanner, based on the device's measured or inferred reference noise pattern, a unique stochastic characteristic of all common digital imaging sensors, including CCD, CMOS (Foveon™ X3), and JFET. The measured or inferred noise pattern of two images may be extracted and then cross correlated, with a high correlation being consistent with a common imager. Various preprocessing techniques may be used to improve tolerance to various types of image transform. It is also possible to perform the analysis without explicit separation of inferred image and inferred noise.