Facial Landmarks Detector Learned by the Structured Output SVM

Abstract

We propose a principled approach to supervised learning of facial landmarks detector based on the Deformable Part Models (DPM). We treat the task of landmarks detection as an instance of the structured output classification. To learn the parameters of the detector we use the Structured Output Support Vector Machines algorithm. The objective function of the learning algorithm is directly related to the performance of the detector and controlled by the userdefined loss function, in contrast to the previous works. Our proposed detector is real-time on a standard computer, simple to implement and easily modifiable for detection of various set of landmarks. We evaluate the performance of our detector on a challenging "Labeled Faces in the Wild" (LFW) database. The empirical results show that our detector consistently outperforms two public domain implementations based on the Active Appearance Models and the DPM. We are releasing open-source code implementing our proposed detector along with the manual annotation of seven facial landmarks for nearly all images in the LFW database.1 There also exists a successful commercial solution OKAO Vision Facial Feature Extraction API (http://www.omron.com) which is used for example in Picasa TM or Apple iPhoto TM software.

Switching Linear Inverse-Regression Model for Tracking Head Pose

“…Moreover, our method also reduces the standard deviation. Compared to the landmark-based method of [39], our method is able to provide an estimation for each test input, whereas the method based on landmarks is unable to provide an output when some of the landmarks are not visible due to extreme head orientations. In this case [39] yields very large errors, e.g.…”

“…The proposed model, referred to as HPE SKF (headpose estimation based on SKF) is compared to the following methods: (i) a landmark-based approach that uses the facial landmark localization method of [39] (Flandmarks) combined with 2D-to-3D landmark-based pose estimation method, namely the PnP (perspective n-point) algorithm available with OpenCV, (ii) a depth head model based on method [26] that learns a 3D head model using 16 manually annotated facial landmarks on multiple frames to learn the model, and ICP (iterative closest point algorithm) to estimate the transformation (rotation and translation) of the head pose between two consecutive frames in order to track the pose over time, (iii) the regression-based method of [10] which is referred to as HPE-GLLiM, and (iv) the regression method [10] combined with a standard Kalman filter [1,6]. Both (i) and (ii) perform tracking.…”

“…Many approaches rely on extracting facial landmarks, on tracking these landmarks over the image sequence and on estimating a rigid transformation between consecutive images, e.g. [13,39], or between consecutive image pairs, e.g. [41].…”

Audio-Video Speaker Diarization for Unsupervised Speaker and Face Model Creation

“…In this paper we use a facial landmarks detector based on the Deformable Part Models [1]. In addition to the detected position of the face, this face detector provides a set of facial landmarks: nose, mouth and canthi corners.…”

Online Speaker Adaptation of an Acoustic Model Using Face Recognition

“…The task of real-time face detection and identification was widely studied and existing solutions are capable to solve this task with high accuracies [8] [9].…”