Tom-vs-Pete Classifiers and Identity-Preserving Alignment for Face Verification

Abstract

We propose a method of face verification that takes advantage of a reference set of faces, disjoint by identity from the test faces, labeled with identity and face part locations. The reference set is used in two ways. First, we use it to perform an "identity-preserving" alignment, warping the faces in a way that reduces differences due to pose and expression but preserves differences that indicate identity. Second, using the aligned faces, we learn a large set of identity classifiers, each trained on images of just two people. We call these "Tom-vs-Pete" classifiers to stress their binary nature. We assemble a collection of these classifiers able to discriminate among a wide variety of subjects and use their outputs as features in a same-or-different classifier on face pairs. We evaluate our method on the Labeled Faces in the Wild benchmark, achieving an accuracy of 93.10%, significantly improving on the published state of the art.

Fisher Vector Faces in the Wild

“…Several papers investigated face descriptors, including LBP and its variants [5,6,12,20,22,25,32,37,38], SIFT [10,20], and learnt representations [25,39]. A particularly interesting idea is to learn and extract semantic face attributes as facial features for identification and other tasks [3,17]. Statistical learning is generally used to map face representations to a final recognition result, with metric or similarity learning being the most popular approach, particularly for AFV [6,10,12,22,40].…”

“…Our first contribution is to show that dense descriptor sampling combined with the improved Fisher Vector (FV) encoding of [24] (Sect. 2) outperforms or performs just as well as the best face verification representations, including the ones that use elaborate face landmark detectors [3,6] and multiple features [12]. The significance of this c 2013.…”

Cross-Age Reference Coding for Age-Invariant Face Recognition and Retrieval

“…Kumar et al [13] propose to use attribute and simile classifier, SVM classifier trained on reference set, for face verification. Berg et al [3] further improve the method by using "Tom-vs-Pete" classifier. Yin et al [39] propose an associate-predict model using 200 identities in Multi-PIE dataset [9] as a reference set.…”

RPM: Random Points Matching for Pair wise Face-Similarity

“…[9] employ a metric learning approach that learns an objective function using a discriminant classifier on a set of positive and negative examples. More recent approaches such as Wolf et al [19] , Yin et al [20] and Berg and Belhumer [4] learn specialized parts classifiers on the pair taking advantage of a reference set of images. An interesting shift from these is the approach used by Kumar et al [11] that learns the attributes called similes (such as hair color, age, gender etc.)…”

“…The current approaches can be largely categorized in two direc-tions; those that compute direct similarity based on proposing new representation methods [1,7,15,18], and those that employ a learning based framework to learn the similarity functions [4,8,9,14,19,20]. Both rely on a point detector to either compute direct similarity by using some feature description method or learn the similarity scores or some parts on the face to later use for recognition.…”