Presentacion DCU
Transcript: Facial Feature Point Detection System Definition Valstar (2008) Geometric Approach based on position, motion, velocity, acceleration and time parameterized features GentleBoosted features Cascade of SVMs Face Detection - Viola & Jones Face Bounding box fitting Reference Points Detection: Eyes Based detection PDM Multi-stage detection Occlusion Management Feature Point and Facial Motion Detection Intra personal Variability Surface Synthesis Correlation between features in the geometric descriptor and the trait labels Sign judgement: analyzes visual changes independently of interpretations; describes physical appearance Vukadinovic and Pantic, Fully Automatic facial feature point detection using Gabor feature based boosted classifiers. IEEE international Conference on Systems, Man and Cybernetics, 2005 Facial Feature Point Detection System Definition Facial Action Coding System (FACS) Action Unit Detection Related Work Frame Based evaluation Evaluation Geometric Approach Appearance Approach Hybrid Approach* Van der Maaten and Hendriks (2011) AAm to extract geometric and appearance features Combined appearance features with SIFT Conditional Random Fields Facial Feature Point Detection Separation of variables required in order to standardize structural information Person Independent framework Leave one subject out scheme 40 subjects with highest number of AU instances Built with (x,y) information of the normalized training samples General Face Detection Elements Facial Action Recognition Cohn Kanade dataset portrays 21 AUs, displayed by over 150 subjects Surface Analysis Structural Capture Conditions Descriptor Generation Two Segments Per frame descriptor Absolute difference between descriptor of frame at time T and that of the initial frame Take advantage of the structure of the point cloud Analyzed contribution of acoustic and visual cues to recognition of personality traits 4 acoustic signals Head, hands and body gestures for visual signals RBF-SVM Extraversion and Locus of Control Facial Expression: change in the face corresponding to reactions against internal of external stimuli Rojas M, Masip D, Vitrià J. Automatic Detection of facial feature points via HOG and Geometric Prior Models, IbPria 2011 Descriptor Analysis - Feature Selection Relates anatomically based motions to facial expressions 1474 dimensional feature vector Koelstra, Pantic and Patras (2010) Motion history map and Quadtree decomposition GentleBoosted features HMM Appearance Search Region Generation Defines all possible facial displays in terms of Action Units (AU), the smallest visible facial movement. Appearance Descriptor Usefulness of facial contextual information 3 stage procedure: Locate facial contextual object of interest Feature point location estimation in rescaled image Refinement of location in rescaled image Face Bounding box fitting process Thank you Feature Extraction Baseline Evaluation Feature Analysis Facial Structure and Social Traits PCA on linguistic judgements of traits identified the dimensions of human facial trait evaluation Developed 2D model of face evaluation: Dominance and Valence Evaluation Results Evaluation of faces is consistent among cultures and influences social processes - Trials Decisions, Political poles. (Oosterhof and Todorov, 2008) Problems Dataset: Behavioral data obtained by Oosterhof and Todorov (2008) Initial Approach: Analyze Descriptors Analyze frame selection impact Multi-Stage Detection Scheme Lepri, Kalimeri and Pianesi (2010) Four Segments S0: Difference between equivalent points in frames at time T and time 0, converted to polar coordinates S1: Difference between each point and all others in the same frame, converted to polar coordinates S2: Previous intra-frame differences compared to the initial frame S3: Intra-frame differences compared to previous frame Social dimensions were modeled by linear changes in the first 50 principal components of the models, bounding the generation to Caucasian faces with no expression. Facial Analysis Analysis of the Receptive Field Rojas M, Masip D, Todorov A, Vitrià J, Automatic Point-based Facial Trait Judgements Evaluation, IEEE CVPR, 2010 Angles were more correlated to trait judgements, suggesting more relevance in symmetry given their distribution in the face (alignment of the eyes in attractiveness, spatial relation of eyes and lips and eyes and nose, mouth position with respect to the chin line in extroversion) Image based Content retrieval Can a relation between social traits and facial characteristics be established? Brahnam (2008) 2880 dimensional feature vector Coordinates of 17 facial feature points Based on HOG applied over the entire region of the face Uses interval between onset and offset Geometric Descriptor Inter personal Variability Frame Based evaluation Change in Descriptor suggests analysis of characteristics Key factor: size and resolution of detector (Eckhardt et al. 2009) MPEG-4 FACS Compute the mean cloud of the data set Defines
Prezi Present
Prezi Video
Prezi Design
