1. Reflection Correspondence for Exposing Photograph Manipulation
Eric Wengrowski^, Zhaohui H. Sun*, Anthony Hoogs*
^ECE, Rutgers University, Piscataway, NJ, USA * Kitware Inc., Clifton Park, NY, USA
Abstract
Reflection Integrity
Experimental Results
Modern photo editing software enables increasingly realistic image manipulations, through splicing, region copy and paste, and content-aware fill. One potential flaw in manipulations is generating realistic object reflections, such as in bodies of water or glass surfaces. Image reflections involve complicated interactions between lighting, surface materials, and geometry, and they can be very hard to fake. Any inconsistency between the directly observed scene and its corresponding reflection is a strong indication of content manipulation. We propose a novel, physical-level image forensic approach to verify the integrity of reflections through identifying mismatches between objects and their reflections. The proposed algorithm consists of reflection-invariant feature detection and matching, geometric transform estimation, and robust change detection between a scene region and its warped reflection region. This approach complements the large body of digital-level forensic methods, with the advantage of being robust to digital-level image transformations such as compression, blurring, and added noise. Experimental results on authentic and manipulated images demonstrate its efficacy.
Robust feature matching between a dinosaur piñata and its mirror reflection, before and after the object refection has been removed
Reflection matching is robust to the choice of reflection masks
Automatic object insertion detection in images with inconsistent reflections
Directly observed scene and indirectly observed reflection have to be consistent in geometry as well as appearance.
The law of reflection. The angle of incidence equals to the angle of reflection, and the incident, normal, and reflected directions are coplanar.
Compression of color. Object reflection appears less colorful than the direct observation. The color may have shifted. However, the area of color region on color gamut does not increase in reflection.
Compression of sharpness. The reflecting surface serves as a low-pass filter and the reflection usually appears blurrier than directly observed scene.
Compression of contrast. Due to directional reflection, refraction and surface abortion, only part of the light reaches the sensor after reflection, resulting in lower contrast.
Processing Modules
Problem & Contributions
Task: predict reflection integrity score and detection mask of the input image
The problem: Assess reflection integrity by predicting reflection locations and geometry using scene context, then comparing reflections to directly imaged objects.
Contributions
Proposed a physical-level image forensic method based on assessing the consistency between scene and reflection
Complementary to the digital-level forensic methods (metadata, JPEG coefficients analysis, PRNU, etc.)
Robust to digital-level manipulations (e.g. spatial transform, compression, and blurring) and digital-analog-digital attacks (e.g. recapture by screen shot, printing, scanning, and projection)
Automatic processing to increase throughput
*James O'Brien and Hany Farid, Exposing photo manipulation with inconsistent reflections, ACM Transactions on Graphics, 2012.
Preprocessing by contrast limited adaptive histogram equalization
Feature point detection in scene and reflection regions
Reflection invariant feature descriptor extraction (ASIFT, MIFT, …)
Descriptor matching by Fast Library of Approximate Nearest Neighbors
Interest point matching by Random Sample Consensus (RANSAC)
Homography estimation
Differencing the directly observed scene region and warped reflection region (e.g. local color standard deviation)
Thresholding and morphological operation on the difference map
Remaining challenges: weak reflections, occlusion, robust change detection
Acknowledgement
This work was supported by AFRL and DARPA under Contract No. FA C Any findings and conclusions or recommendations expressed in this material are solely the responsibility of the authors and do not necessarily represent the official views of AFRL, DARPA, or the U.S. Government.