Presentation on theme: "Exposing Photo Manipulation with Inconsistent Reections James F. OBrien University of California, Berkeley Hany Farid Dartmouth College Presented by: Jinjie."— Presentation transcript:
Exposing Photo Manipulation with Inconsistent Reections James F. OBrien University of California, Berkeley Hany Farid Dartmouth College Presented by: Jinjie Lin All images presented in this document is from the paper except the ones whose source link are provided.
Where can we get the lastet SIG papers?
Photographs can no longer be trusted!
Is this a real photo? A forged image
Is this a real photo?
Is it really so easy?
Is this a real photo?
Digital watermarking Embedding information into a digital signal which may be used to verify its authenticity or the identity of its owners Drawback – Watermark must be inserted at the time of recording
Passive forensic techniques Operate in the absence of watermarks or signatures. Assumption: – Although photo manipulation may leave no obvious visual clues, it may alter some geometric or statistical property in the image.
Exposing Photo Manipulation with Inconsistent Reections
Planar reflection geometry: The reflection Vanishing Point Observation: -PR must be perpendicular to MB -Photo: liner perspective projection -Parallel lines must all converge to the same vanishing point.
Planar reflection geometry: Reflection Line Midpoints Observation: The reflection line midpoint must lie in the plane of the reflecting surface.
Planar reflection geometry: Reflection Line Midpoints p r Test: -for a line pr, check whether it converge to the same vanishing point v; -Compute the midpoint, m; -Compare m to other scene elements for consistency.
Planar reflection geometry: The Center of Projection Observation: For a given image, the center of projection is unique. /jiaocai.php?bookpage=7_d
How to find out the center of projection from an image When vanishing points for three mutually orthogonal directions are known. Common in man-made environments
How to find out the center of projection from images The computation is extremely unstable. Instead, they compute a cloud of plausible locations rather than a single location.
Detecting Inconsistencies Three tests for exposing inconsistencies in an image – ill-defined reflection vanishing points – Implausible midpoints – Inconsistent centers of projection Not all of these tests are applicable to any given image. – Require some human understanding of the image content.
Results lack of a well-dened reection vanishing point
each object has a well-dened reection vanishing point, but they are mutually inconsistent.
Blue lines: scene features with their corresponding reections in the building windows. Green: linear features that should be perpendicular to the building front
head has been removed the height of her hair has been altered.
Conclusion Present a set of test to expose photo manipulation. Require the input photo should contain planar reectors where parts of the scene can be observed both directly and indirectly through reection. Is not able to prove a photo is authentic, but it can prove a photo is fake.