Citation Context

...arance models have become increasingly prominent in computer vision and graphics. Parameterized Appearance Models (PAMs) have been proven useful for alignment, detection, tracking, and face synthesis =-=[24, 25, 26, 27]-=-. In particular, Active Appearance Models (AAMs) have proven an excellent tool for detecting and aligning facial features. AAMs [28, 29] typically fit their shape and appearance components to an image...

Citation Context

...ased technician productivity by automating certain features of their work, making particularly tedious or time consuming tasks in image editing simpler and quicker. For example, the ”magic wand” tool =-=[5]-=- used to automatically select image regions based on pixel to pixel similarity, or more recently, the use of segmentation algorithms for user-assisted foreground/background separation and matting. Sim...

Citation Context

...arance models have become increasingly prominent in computer vision and graphics. Parameterized Appearance Models (PAMs) have been proven useful for alignment, detection, tracking, and face synthesis =-=[24, 25, 26, 27]-=-. In particular, Active Appearance Models (AAMs) have proven an excellent tool for detecting and aligning facial features. AAMs [28, 29] typically fit their shape and appearance components to an image...

Citation Context

...e parameter C adjusts the importance of the constraints in the minimization. Eq. (1) can be solved with semi-definite programming approaches. We use the CVX package for disciplined convex programming =-=[33, 34]-=-. Restricting A to be diagonal is equivalent to individually weighting the features. While a full matrix could be used, in our experience diagonal and full matrices provide comparable performance. Bec...

Citation Context

...e system agree. Concurrent validity refers to whether there is agreement between systems. Percent agreement was computed using percentage agreement Ekman & Frisen [40], and as a coefficient Kappa (k) =-=[41]-=-. Kappa is a more rigorous metric as it is controls for agreements due to chance. Agreement is reported for both all intensity levels and for intensity levels B and higher. In the original FACS manual...

Citation Context

...onset and offset estimation. A large literature addresses supervised, unsupervised, and semisupervised metric learning, see [31] for an extensive review. In this paper, we will follow the approach of =-=[32]-=- and frame the metric learning problem as a semidefinite programming (SDP) optimization. Suppose that d features for each frame i are stacked into a vector, fi ∈ ℜ d×1 . Let f p i denote a frame i wit...

Citation Context

...ted strong concurrent validity with manual FACS coding. Index Terms—Action Unit Recognition, Facial Expression Analysis, Facial Action Coding System I. INTRODUCTION FACS (Facial Action Coding System: =-=[1]-=-) coding is the state of the art in manual measurement of facial action. FACS coding, however, is labor intensive and difficult to standardize. A goal of automated FACS Manuscript received July 9, 200...

Citation Context

...arance models have become increasingly prominent in computer vision and graphics. Parameterized Appearance Models (PAMs) have been proven useful for alignment, detection, tracking, and face synthesis =-=[24, 25, 26, 27]-=-. In particular, Active Appearance Models (AAMs) have proven an excellent tool for detecting and aligning facial features. AAMs [28, 29] typically fit their shape and appearance components to an image...

Citation Context

... for alignment, detection, tracking, and face synthesis [24, 25, 26, 27]. In particular, Active Appearance Models (AAMs) have proven an excellent tool for detecting and aligning facial features. AAMs =-=[28, 29]-=- typically fit their shape and appearance components to an image through a gradient descent, although other optimization approaches have been employed with similar results. Figure 1 shows how a person...

Citation Context

...lly weighting the features. While a full matrix could be used, in our experience diagonal and full matrices provide comparable performance. Because a diagonal matrix results in a simpler minimization =-=[32, 35]-=-, we use this approach for the results presented in Section I IV. D. Graphical user interface There exist several commercial packages for manual event coding. These systems are general purpose and mus...

Citation Context

...eams on automatic analysis of facial expression consider emotion-specified expressions (e.g., happy or sad) and anatomically based facial actions (e.g., FACS). The pioneering work of Black and Yacoob =-=[10]-=- recognized facial expressions by fitting local parametric motion models to regions of the face and then feeding the resulting parameters to a nearest neighbor classifier for expression recognition. D...

Citation Context

...tection This section describes the procedure to learn an optimal metric for onset and offset estimation. A large literature addresses supervised, unsupervised, and semisupervised metric learning, see =-=[31]-=- for an extensive review. In this paper, we will follow the approach of [32] and frame the metric learning problem as a semidefinite programming (SDP) optimization. Suppose that d features for each fr...

Citation Context

... smiling), from onset, peak, to offset, with superimposed Active Appearance Model (AAM) mesh. Using FastFACS only the peak needs to be labeled and the onset/offset are estimated automatically. coding =-=[2, 3, 4]-=- is to eliminate the need for manual coding and realize automatic recognition and analysis of facial actions.Success of this effort depends on access to reliably coded corpora of FACS-coded images fro...

Citation Context

... construct a non-linear manifold that factorizes identity from expression. Recently there has been an emergence of efforts toward explicit automatic analysis of facial expressions into elementary AUs =-=[13, 8, 14, 15, 16, 17, 18, 19]-=-. The most relevant work is the one that addresses the temporal segmentation of Action units into onset, offset, and peak. Pantic and Pantras [4] used a rule-basedJOURNAL OF LATEX CLASS FILES, MAY 20...

Citation Context

... construct a non-linear manifold that factorizes identity from expression. Recently there has been an emergence of efforts toward explicit automatic analysis of facial expressions into elementary AUs =-=[13, 8, 14, 15, 16, 17, 18, 19]-=-. The most relevant work is the one that addresses the temporal segmentation of Action units into onset, offset, and peak. Pantic and Pantras [4] used a rule-basedJOURNAL OF LATEX CLASS FILES, MAY 20...

Citation Context

... smiling), from onset, peak, to offset, with superimposed Active Appearance Model (AAM) mesh. Using FastFACS only the peak needs to be labeled and the onset/offset are estimated automatically. coding =-=[2, 3, 4]-=- is to eliminate the need for manual coding and realize automatic recognition and analysis of facial actions.Success of this effort depends on access to reliably coded corpora of FACS-coded images fro...

Citation Context

... construct a non-linear manifold that factorizes identity from expression. Recently there has been an emergence of efforts toward explicit automatic analysis of facial expressions into elementary AUs =-=[13, 8, 14, 15, 16, 17, 18, 19]-=-. The most relevant work is the one that addresses the temporal segmentation of Action units into onset, offset, and peak. Pantic and Pantras [4] used a rule-basedJOURNAL OF LATEX CLASS FILES, MAY 20...

Citation Context

...frame j using an affine transform intended to remove the rigid movement of the head while retaining the elastic deformations of facial actions. Other geometric normalizations may be used (e.g., 2D+3D =-=[30]-=-) depending on the extend of non-rigid head rotation present in the data. Once both frames are commonly referenced, the landmarks are stacked into vectors fi and fj, and kij = e −d(fi ,fj ) 2σ2 Fig. 3...

Citation Context

...oding. Many AU fade slowly as the face relaxes, which attenuates the observable signal values of offsets. Indeed, only in some selected types of smiles (e.g., embarrassed), does one find fast offsets =-=[43]-=-. Perhaps the most important confound for AU 12 is mouth movement due to speech, which makes similarity based methods fail. C. Efficiency of Fast-FACS Fast-FACS reduced total coding time by one half o...

Citation Context

... offset. Koelstra and Pantic [20] used Hidden Markov Models to detect the onset and offset in AUs. In their case, they considered the apex to be frame in the middle of the segment. Valstar and Pantic =-=[21]-=- combined Hidden Markov Models and Support Vector Machines to model the temporal dynamics of facial actions. They considered the onset, apex, and offset frames as different classes. Accuracy was measu...

Citation Context

... construct a non-linear manifold that factorizes identity from expression. Recently there has been an emergence of efforts toward explicit automatic analysis of facial expressions into elementary AUs =-=[13, 8, 14, 15, 16, 17, 18, 19]-=-. The most relevant work is the one that addresses the temporal segmentation of Action units into onset, offset, and peak. Pantic and Pantras [4] used a rule-basedJOURNAL OF LATEX CLASS FILES, MAY 20...

Citation Context

...c and Euclidean distance. Temporal error for onsets was relatively low. The mean error ranged from 3.8 to 13.1 for AUs (1,2,4,10,15). This is within the standard of acceptable error for manual coding =-=[42]-=-. Inter-system agreement for offsets was lower and more variable. With the exception of AU 12, mean offset error ranged from 10.6 to 36.2 frames. Mean onset error for AU 12 was 32.13 frames, mean offs...

Citation Context

...s to whether two coders using the same system agree. Concurrent validity refers to whether there is agreement between systems. Percent agreement was computed using percentage agreement Ekman & Frisen =-=[40]-=-, and as a coefficient Kappa (k) [41]. Kappa is a more rigorous metric as it is controls for agreements due to chance. Agreement is reported for both all intensity levels and for intensity levels B an...

Citation Context

...ther automatically estimated. In calculating inter-coder and inter-system agreement, a window of agreement of ±.5 seconds (15 frames) was used. In FACS coding, it is typical allowed a margin of error =-=[38]-=-. Inter-coder agreement [39] refers to whether two coders using the same system agree. Concurrent validity refers to whether there is agreement between systems. Percent agreement was computed using pe...

Citation Context

.... In calculating inter-coder and inter-system agreement, a window of agreement of ±.5 seconds (15 frames) was used. In FACS coding, it is typical allowed a margin of error [38]. Inter-coder agreement =-=[39]-=- refers to whether two coders using the same system agree. Concurrent validity refers to whether there is agreement between systems. Percent agreement was computed using percentage agreement Ekman & F...

Citation Context

...at is comparable to commercially available packages in easy of use, while enabling fast reliable coding. Figure 1 shows the main idea of the paper. Preliminary versions of this work were presented in =-=[6, 7]-=-. Section II-A reviews previous work on FACS. Section III describes the proposed procedure to improve speed and reduce variability in FACS coding. Section III-D describes advances in the graphical use...

Citation Context

... construct a non-linear manifold that factorizes identity from expression. Recently there has been an emergence of efforts toward explicit automatic analysis of facial expressions into elementary AUs =-=[13, 8, 14, 15, 16, 17, 18, 19]-=-. The most relevant work is the one that addresses the temporal segmentation of Action units into onset, offset, and peak. Pantic and Pantras [4] used a rule-basedJOURNAL OF LATEX CLASS FILES, MAY 20...

Citation Context

...arance models have become increasingly prominent in computer vision and graphics. Parameterized Appearance Models (PAMs) have been proven useful for alignment, detection, tracking, and face synthesis =-=[24, 25, 26, 27]-=-. In particular, Active Appearance Models (AAMs) have proven an excellent tool for detecting and aligning facial features. AAMs [28, 29] typically fit their shape and appearance components to an image...

Citation Context

... smiling), from onset, peak, to offset, with superimposed Active Appearance Model (AAM) mesh. Using FastFACS only the peak needs to be labeled and the onset/offset are estimated automatically. coding =-=[2, 3, 4]-=- is to eliminate the need for manual coding and realize automatic recognition and analysis of facial actions.Success of this effort depends on access to reliably coded corpora of FACS-coded images fro...

Citation Context

...(AU6). FACS is recognized as the most comprehensive and objective means for measuring facial movement currently available, and it has become the standard for facial measurement in behavioral research =-=[8, 9]-=-. Human-observer-based methods like FACS are time consuming to learn and use, and they are difficult to standardize, especially across laboratories and over time. A goal of automated FACS coding [2, 3...

Citation Context

... to a nearest neighbor classifier for expression recognition. De la Torre et al. [11] used condensation and appearance models to simultaneously track and recognize facial expression. Lee and Elgammal =-=[12]-=- used multi-linear models to construct a non-linear manifold that factorizes identity from expression. Recently there has been an emergence of efforts toward explicit automatic analysis of facial expr...

Citation Context

...entation of Action units into onset, offset, and peak. Pantic and Pantras [4] used a rule-basedJOURNAL OF LATEX CLASS FILES, MAY 2009 3 method to separate onset, apex and offset. Koelstra and Pantic =-=[20]-=- used Hidden Markov Models to detect the onset and offset in AUs. In their case, they considered the apex to be frame in the middle of the segment. Valstar and Pantic [21] combined Hidden Markov Model...

Citation Context

...III-D describes the graphical user interface (GUI), that also leads to improvements in speed of coding. A. Active Appearance Tracking There exist a variety of methods for facial feature tracking. See =-=[23]-=- for a review. Over the last decade, appearance models have become increasingly prominent in computer vision and graphics. Parameterized Appearance Models (PAMs) have been proven useful for alignment,...

Citation Context

...(AU6). FACS is recognized as the most comprehensive and objective means for measuring facial movement currently available, and it has become the standard for facial measurement in behavioral research =-=[8, 9]-=-. Human-observer-based methods like FACS are time consuming to learn and use, and they are difficult to standardize, especially across laboratories and over time. A goal of automated FACS coding [2, 3...

Citation Context

... construct a non-linear manifold that factorizes identity from expression. Recently there has been an emergence of efforts toward explicit automatic analysis of facial expressions into elementary AUs =-=[13, 8, 14, 15, 16, 17, 18, 19]-=-. The most relevant work is the one that addresses the temporal segmentation of Action units into onset, offset, and peak. Pantic and Pantras [4] used a rule-basedJOURNAL OF LATEX CLASS FILES, MAY 20...

Citation Context

... expressions by fitting local parametric motion models to regions of the face and then feeding the resulting parameters to a nearest neighbor classifier for expression recognition. De la Torre et al. =-=[11]-=- used condensation and appearance models to simultaneously track and recognize facial expression. Lee and Elgammal [12] used multi-linear models to construct a non-linear manifold that factorizes iden...

Citation Context

...of fully automated expression or AU coding. More recently, two groups have proposed hybrid systems that make use of more unsupervised learning techniques to augment manual coding of AUs. Zhang et al. =-=[22]-=- proposed an active learning approach to improve speed and accuracy in AU labeling. In their approach, a sequence is labeled with an automatic system, and a user then is asked to label the frames that...

Citation Context

...s it substantially reduce the time required to complete FACS coding? To pursue these questions, we conducted several experiments using a relatively challenging corpus of FACS coded video, the RU-FACS =-=[36]-=- video data-set. RUFACS is a false opinion dataset recorded at Rutgers University. It consists of non-posed facial behavior of 100 participants who were observed for approximately 2.5 minutes each whi...

Citation Context

... for alignment, detection, tracking, and face synthesis [24, 25, 26, 27]. In particular, Active Appearance Models (AAMs) have proven an excellent tool for detecting and aligning facial features. AAMs =-=[28, 29]-=- typically fit their shape and appearance components to an image through a gradient descent, although other optimization approaches have been employed with similar results. Figure 1 shows how a person...