an image-based approach to video copy detection with spatio -temporal post-filtering
DESCRIPTION
An Image-Based Approach to Video Copy Detection With Spatio -Temporal Post-Filtering. Matthijs Douze , Hervé Jégou , and Cordelia Schmid , Senior Member, IEEE. INTRODUCTION. Common distortions are 1. scaling 2. compression 3. cropping 4. camcording. FRAME INDEXING (step1~6). - PowerPoint PPT PresentationTRANSCRIPT
An Image-Based Approach to Video Copy Detection With
Spatio-Temporal Post-Filtering
Matthijs Douze, Hervé Jégou, and Cordelia Schmid, Senior
Member, IEEE
INTRODUCTION
Common distortions are 1. scaling2. compression 3. cropping4. camcording
FRAME INDEXING (step1~6)
a. Frame Sampling
1. Uniform sampling
2. Keyframes
b. Local Features (salient interest
points)
invariant :
1. Scale change
2. Image rotation
3. Noise
c. Bag-of-Features and Hamming
Embedding
SPATIO-TEMPORAL VERIFICATION
A. Spatio-Temporal Transformation
B. Temporal GroupingC. Spatial Verification(next)D.Score Aggregation
Strategy
Spatial Verification 1. take all point matches from the matching frames. 2. estimate possible similarity transformations from all matching points with a Hough transform.(next)
3. compute and score possible affine transformations. 4. select the maximum score over all possible hypotheses.
ExperimentA. Parameter Optimization(next)B. Handling of Trecvid AttacksC. Trecvid Copy Detection Results
Conclusion
Our video copy detection system outperforms other submitted results on all transformations. This is due to a very accurate image-level matching. Run KEYSADVES, which is more scalable, shows that our system still obtains excellent results with a memory footprint and query time reduced 20 times.