Gabor-based gradient orientation pyramid for kinship verification under uncontrolled environments

Visual Transformation Aided Contrastive Learning for Video-Based Kinship Verification

“…Then, several studies have aimed to engineer powerful facial appearance representations such as Spatial Pyramid LEarningbased descriptors [38], DAISY descriptors [12], Gaborbased Gradient Orientation Pyramid [39], Self Similarity Representation [16], semantic-related attributes [32], SIFT flow based genetic Fisher vector feature [26], etc.…”

Kinship verification in multi-linear coherent spaces

“…Different approaches roughly fall into two categories, either focusing on metric learning for more accurate distance computation [29,32,34,40], or feature engineering for better appearance encoding [41,47,48]. To obtain more robust measure, a recent method dubbed neighborhood repulsed metric learning (NRML) was proposed in [32] and [29].…”

“…Zhou et al [48] proposed a Gabor-based gradient orientation pyramid feature and used it with Support Vector Machine (SVM) classifier for kinship verification. In [41], instead of using low-level hand-crafted features, mid-level features were computed based on the decision values from an SVM classifier trained with an independent face dataset without kinship relation labels.…”

Kinship verification in the wild: The first kinship verification competition

“…Over the past five years, several kinship verification via face images approaches have been proposed in computer vision and biometrics [11,35,31,39,36,18,40,23,10,25]. Generally, these methods can be categorized into two streams: descriptor-based [11,39,40,18] and similarity learning-based [35,36,31,25].…”

“…Generally, these methods can be categorized into two streams: descriptor-based [11,39,40,18] and similarity learning-based [35,36,31,25]. For descriptor-based methods, some important cues such as skin color [11], histogram of gradient [11], Gabor gradient orientation pyramid [40], salient part [23], self-similarity [18], and dynamic expressions [10], are usually employed for face representation. For similarity learning-based methods, subspace and metric learning are used to learn a semantic feature space to better measure the similarities of face samples.…”