Stephan Tschechne Chair for Image Understanding Computer Science Technische Universität München Designing vs. Learning the Objective Function for Face Model Fitting Abschlußvortrag Diplomarbeit
/15 Technische Universität München Stephan Tschechne Model-based Image Understanding Face Model Fitting Objective Functions Experimental Results Overview:
/15 Technische Universität München Stephan Tschechne Understanding Facial Images Various Applications Identification Mimics Hands-free Control Image Database: 850 Natural Images
/15 Technische Universität München Stephan Tschechne Deformable Face Model 134 Contour Points Perform PCA Point Distribution Model Description of an Instance: Parameter Vector p = (x,y,scaling,rotation,deform1..deform17)
/15 Technische Universität München Stephan Tschechne Objective Function Fitting Algorithms Search for Correct p: Optimisation Problem Objective Function Calculates Fitting Accuracy Lowest Value for Correct Solution F(Img,p1)=0.0 F(Img,p2)=0.3F(Img,p3)=0.6
/15 Technische Universität München Stephan Tschechne Requirements Formulation of Requirements for Robust Objective Functions: R1: Correct Position of Minimum R2: One Minimum R3: Continuous Behaviour R4: Gradient Vectors Point Away Optimal Objective Functions:
/15 Technische Universität München Stephan Tschechne Traditional Objective Functions Calculation of Objective Function Value ? Intuitive approach: Manual Selection of Salient Features: Distance to Edges..
/15 Technische Universität München Stephan Tschechne Traditional Objective Functions …or Distance to Edges from Skin Colour Images
/15 Technische Universität München Stephan Tschechne Traditional Approach Problem: Desired Edges are not the Strongest Ones
/15 Technische Universität München Stephan Tschechne Contribution Robust Objective Function Better Fulfillment of the Requirements ?!
/15 Technische Universität München Stephan Tschechne Learning the Robust Objective Function Training data: Ground Truth from Image Database Haar-like Features Desired Value from Optimal Objective Function Machine Learns Rules with Model Trees
/15 Technische Universität München Stephan Tschechne Training Data: Feature Values F i Result R Deliberately Move Instance to Gather Values Model Trees Learn: F(Feature Values) Result F 1 =134 F 2 =66 … R=0.7 F 1 =54 F 2 =234 … R=0.0 F 1 =281 F 2 =11 … R=0.5
/15 Technische Universität München Stephan Tschechne Experimental Results Center: Correct Parameters Axes: Variation of p towards …...translation..deformation
/15 Technische Universität München Stephan Tschechne Challenges: Image Database with Natural Images Database High Dimensionality of Parameter Vector Verification of Requirements Future research: Model Tracking Other Models: 3D, Appearance Models.. Different Features Other Positions for Features
/15 Technische Universität München Stephan Tschechne The End. Any Questions ?