Multi-view face alignment using direct appearance models

Abstract

Alignment makes face distribution statistically more compact than un-aligned faces and provides a good basis for face modeling, recognition and synthesis. In this paper, we present a method for multi-view face alignment using a new model called direct appearance model (DAM). Like active appearance model (AAM), DAM also makes use of both shape and texture constraints; however, it does this without combining shape and texture as in AAM. The way that DAM models shapes and textures has the following advantages as compared to AAM: (1) DAM subspaces include admissible appearances previously unseen in AAM, (2) the convergence and accuracy are improved, and (3) the memory requirement is cut down to a large extent. Extensive experiments are presented to evaluate the DAM alignment in comparison with AAM.

Structure-perceptron learning of a hierarchical log-linear model

“…The dataset [22] contains ground truth of standard 65 key points which lie along the boundaries of face components with semantic meaning, i.e, eyes, nose, mouth and cheek. We use part of this dataset for training (200 images) and part for testing (80 images).…”

Face Alignment Under Various Poses and Expressions

“…Different from the view ranges presented in [6] In the system illustrated in Fig. 1, right now 'Facial landmark extraction' [11] is implemented for frontal faces and 'Pose estimation' [12] …”

“…In this paper, we focus our work on ASM. In recent years, many new derivative methods have been proposed, such as that of ASM-based, TC-ASM [3], W-ASM [4], and Haar-wavelet ASM [5], that of AAM-based, DAM [6], AWN [7]. However the problem is still an unsolved one for practical applications since their performances are very sensitive to large variations in face pose and especially in face expression although usually they can acquire good results on neutral faces, which may be caused by the global shape model that is not so powerful to represent changes in face components under complex pose and expression variations.…”

Learning a Hierarchical Deformable Template for Rapid Deformable Object Parsing

“…We show that they are very effective in terms of performance and speed (roughly 20 seconds for a typical 300 × 200 image -the speed increases approximately linearly in the size of the image) when evaluated on large datasets which include horses [16] and cows [17]. In particular, to illustrate versatility, we demonstrate state-of-the-art results for different tasks such as object segmentation (evaluated on the Weizmann horse dataset [16]) and matching/alignment (evaluated on the face dataset - [18], [19]). The results on the alignment task on the face dataset are particularly interesting because we are comparing to results obtained by methods such as Active Appearance Models [20] which are specialized for faces and which have been developed over a period of many years (while we spent one week in total to run this application including the time to obtain the dataset).…”