200 likes | 294 Views
Classifying Covert Photographs. CVPR 2012 POSTER. Outline. Introduction Combine Image Features and Attributes Experiment Conclusion. Introduction. Why doing this classification ? Image/video acquisition devices New Internet technologies What is covert? Secret photography.
E N D
Classifying Covert Photographs CVPR 2012 POSTER
Outline • Introduction • Combine Image Features and Attributes • Experiment • Conclusion
Introduction • Why doing this classification? • Image/video acquisition devices • New Internet technologies • What is covert? • Secret photography
Introduction • Challenges • Database construction • Training set covert:1200 regular:4800 • Testing set covert:300 regular:1200 • Attribute annotation
Combine Image Features and Attributes • Low-Level Image Features • Bag of Features(BoF) • Color GIST • Color moments • Edge Orientation Histogram • Gray Histogram
Combine Image Features and Attributes • Low-Level Image Features • Gray Level Co-occurrence Matrix • Hue descriptor • Local Binary Pattern • Pyramid histogram of orientation gradient • Spatiogram
Combine Image Features and Attributes • Attribute Classifiers and Attribute Features
Combine Image Features and Attributes • Fusion with Multiple Kernels Learning(MKL)
Combine Image Features and Attributes • Fusion with Multiple Kernels Learning(MKL) • Feature normalization and kernel standardization
Experiment • Performance evaluation metrics • AUC • 1-EER
Experiment • Evaluation of MKL algorithm
Experiment • Evaluation of MKL algorithm
Experiment • Evaluation of MKL algorithm
Experiment • Evaluation of MKL algorithm
Experiment • Evaluation of MKL algorithm
Experiment • Evaluation of MKL algorithm
Conclusion • Appropriate features are really important to the accuracy. • Multiple Kernel Learning