Image re-sampling detection through a novel interpolation kernel

Area of Science:

• Digital Image Processing
• Computer Vision
• Signal Processing

Background:

• Image re-sampling is fundamental to digital image manipulation.
• Traces of re-sampling transformations are detectable.
• Existing interpolation kernels have limitations.

Purpose of the Study:

• Propose a new re-sampling interpolation kernel with controllable parameters.
• Demonstrate its ability to mimic common interpolation kernels.
• Develop a method to characterize and detect correlation coefficients in re-sampling.

Main Methods:

• Introduced a novel interpolation kernel with five independent parameters.
• Utilized gradient-based minimization of an error function.
• Assessed the method on a database of 11,000 re-sampled images.

Main Results:

• The proposed kernel successfully imitates existing interpolation kernels.
• The method effectively characterizes and detects re-sampling correlation coefficients.
• Achieved better performance and reduced processing time compared to a reference method.

Conclusions:

• The novel interpolation kernel is effective for detecting image re-sampling.
• The proposed method offers a robust and efficient solution for image forensics.
• Validated suitability for complex image transformations.