Face Model Fitting with Generic, Group-specific, and Person- specific Objective Functions Chair for Image Understanding and Knowledge-based Systems Institute for Informatics Technische Universität München Sylvia Pietzsch
2008, January 22 nd 2/15 Technische Universität München Sylvia Pietzsch Overview Model-based Image Interpretation Generic Objective Functions Specific Objective Functions Experimental Results Conclusion and Outlook
2008, January 22 nd 3/15 Technische Universität München Sylvia Pietzsch Model-based image interpretation The model The model contains a parameter vector that represents the model’s configuration. The objective function Calculates a value that indicates how accurately a parameterized model matches an image. The fitting algorithm Searches for the model parameters that describe the image best, i.e. it minimizes the objective function.
2008, January 22 nd 4/15 Technische Universität München Sylvia Pietzsch Local Objective Functions
2008, January 22 nd 5/15 Technische Universität München Sylvia Pietzsch Ideal Objective Functions P1:Correctness property: Global minimum corresponds to the best fit. P2:Uni-modality property: The objective function has no local extrema. ¬ P1 P1 ¬P2 P2 Don’t exist for real-world images Only for annotated images: f n ( I, x ) = | c n – x |
2008, January 22 nd 6/15 Technische Universität München Sylvia Pietzsch Learning the Objective Function x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x Ideal objective function generates training data Machine Learning technique generates calculation rules
2008, January 22 nd 7/15 Technische Universität München Sylvia Pietzsch Learning the Objective Function (1)
2008, January 22 nd 8/15 Technische Universität München Sylvia Pietzsch Learning the Objective Function (2)
2008, January 22 nd 9/15 Technische Universität München Sylvia Pietzsch Learning the Objective Function (3) 6 styles · 3 sizes · (5 · 5) locations = 450 features
2008, January 22 nd 10/15 Technische Universität München Sylvia Pietzsch Specific Objective Functions Single Images The objective function has to take any appearance of a human face into consideration. ➱ moderate accuracy Image Sequence The appearance of a person‘s face only changes slightly. Consider particular characteristics of the visible person, e.g. beard, glasses, bald head,... ➱ increase of accuracy Challenges: Learn specific objective functions for groups of persons offline. Detect the correct group online.
2008, January 22 nd 11/15 Technische Universität München Sylvia Pietzsch Evaluation: Fitting Results 45 persons from news broadcasts on TV
2008, January 22 nd 12/15 Technische Universität München Sylvia Pietzsch Group-specific Objective Functions automatically subdivide persons into partitions learn objective function for each partition best possible partitioning: minimize the mean fitting error when applying the group-specific objective functions to the partitions number of partitions influences specificity
2008, January 22 nd 13/15 Technische Universität München Sylvia Pietzsch Experimental Results P26 P44 P42 λ = 3.9
2008, January 22 nd 14/15 Technische Universität München Sylvia Pietzsch Conclusion and Outlook large variation in facial appearance challenges model fitting trade-off between generality and accuracy group-specific objective functions perform substantially better for persons within the group of avail for tracking face models through image sequences Further work: automatic classification to determine on-line, which partition a person belongs to
2008, January 22 nd 15/15 Technische Universität München Sylvia Pietzsch Thank you!