Siyu Xia

Understanding Kin Relationships in a Photo

There is an urgent need to organize and manage images of people automatically due to the recent explosion of such data on the Web in general and in social media in particular. Beyond face detection and face recognition, which have been extensively studied over the past decade, perhaps the most interesting aspect related to human-centered images is the relationship of people in the image. In this work, we focus on a novel solution to the latter problem, in particular the kin relationships. To this end, we constructed two databases: the first one named UB KinFace Ver2.0, which consists of images of children, their young parents and old parents, and the second one named FamilyFace. Next, we develop a transfer subspace learning based algorithm in order to reduce the significant differences in the appearance distributions between children and old parents facial images. Moreover, by exploring the semantic relevance of the associated metadata, we propose an algorithm to predict the most likely kin relationships embedded in an image. In addition, human subjects are used in a baseline study on both databases. Experimental results have shown that the proposed algorithms can effectively annotate the kin relationships among people in an image and semantic context can further improve the accuracy.

Kinship verification through transfer learning

Because of the inevitable impact factors such as pose, expression, lighting and aging on faces, identity verification through faces is still an unsolved problem. Research on biometrics raises an even challenging problem--is it possible to determine the kinship merely based on face images? A critical observation that faces of parents captured while they were young are more alike their childrens compared with images captured when they are old has been revealed by genetics studies. This enlightens us the following research. First, a new kinship database named UB KinFace composed of child, young parent and old parent face images is collected from Internet. Second, an extended transfer subspace learning method is proposed aiming at mitigating the enormous divergence of distributions between children and old parents. The key idea is to utilize an intermediate distribution close to both the source and target distributions to bridge them and reduce the divergence. Naturally the young parent set is suitable for this task. Through this learning process, the large gap between distributions can be significantly reduced and kinship verification problem becomesmore discriminative. Experimental results show that our hypothesis on the role of young parents is valid and transfer learning is effective to enhance the verification accuracy.

Genealogical face recognition based on UB KinFace database

In this paper, we consider a challenging problem raised in biometric recently, genealogical face recognition. Practically, kinship can be proved via several methods, i.e., gene match. We argue in this paper that merely based on facial appearance can we present an accepted result on kinship verification, though there is remarkable appearance variance between kinship members. Essentially, as an extension of our former study, we proceed with the following work. First, an extended kinship database named “UB KinFace Ver2.0” is introduced. Second, we propose to use robust local Gabor filters to extract genetic-invariant features. Finally, metric learning and transfer subspace learning are adopted to abridge the great discrepancy between children and their old parents. The key idea is to take advantage of some intermediate data, e.g. young parents, close to both source and target data. Experimental results demonstrate the effectiveness of our kinship verification methods.

Tree-Structured Support Vector Machines for Multi-class Classification

In this paper, a non-balanced binary tree is proposed for extending support vector machines (SVM) to multi-class problems. The non-balanced binary tree is constructed based on the prior distribution of samples, which can make the more separable classes separated at the upper node of the binary tree. For an kclass problem, this method only needs k-1 SVM classifiers in the training phase, while it has less than kbinary test when making a decision. Further, this method can avoid the unclassifiable regions that exist in the conventional SVMs. The experimental result indicates that maintaining comparable accuracy, this method is faster than other methods in classification.

Identity and Kinship Relations in Group Pictures

This chapter studies the problem of identifying people in group pictures. That is, determining from a gallery of people who appear in a given picture. This is a well-studied problem that is becoming increasingly important given the recent explosion in usage of social networks. In this chapter we make two distinct contributions to this problem. First, we use novel kinship similarity to make better estimation of identity. Specifically, we use unary costs based on state-of-the-art face recognition algorithms and as pairwise cost we use the kinship similarity of the people in the image. Second, with these values we formulate a collection-specific MRF MAP estimation (labelling) problem and use existing MRF MAP estimation methods to solve it. To evaluate the proposed method, a family photo database is collected from the Internet. Experiments show that for group pictures of family members (family pictures) our method obtains the state-of-the-art performance, while performing competitively in nonfamily group pictures.

Context-based embedded image compression using binary wavelet transform

Binary wavelet transform (BWT) has several distinct advantages over the real wavelet transform (RWT), such as the conservation of alphabet size of wavelet coefficients, no quantization introduced during the transform and the simple Boolean operations involved. Thus, less coding passes are engaged and no sign bits are required in the compression of transformed coefficients. However, the use of BWT for the embedded grayscale image compression is not well established. This paper proposes a novel Context-based Binary Wavelet Transform Coding approach (CBWTC) that combines the BWT with a high-order context-based arithmetic coding scheme to embedded compression of grayscale images. In our CBWTC algorithm, BWT is applied to decorrelate the linear correlations among image coefficients without expansion of the alphabet size of symbols. In order to match up with the CBWTC algorithm, we employ the gray code representation (GCR) to remove the statistical dependencies among bi-level bitplane images and develop a combined arithmetic coding scheme. In the proposed combined arithmetic coding scheme, three highpass BWT coefficients at the same location are combined to form an octave symbol and then encoded with a ternary arithmetic coder. In this way, the compression performance of our CBWTC algorithm is improved in that it not only alleviate the degradation of predictability caused by the BWT, but also eliminate the correlation of BWT coefficients in the same level subbands. The conditional context of the CBWTC is properly modeled by exploiting the characteristics of the BWT as well as taking advantages of non-causal adaptive context modeling. Experimental results show that the average coding performance of the CBWTC is superior to that of the state-of-the-art grayscale image coders, and always outperforms the JBIG2 algorithm and other BWT-based binary coding technique for a set of test images with different characteristics and resolutions.

Online Real Adaboost with Co-training for Object Tracking

One of the major challenges of object tracking is to tackle appearance variations, possibly caused by the change of object postures, size, and occlusions. In this paper an adaptive tracking system is presented, which integrates online semisupervised classification and particle filter efficiently. To identify object pixels from background accurately, classifiers are trained online using real Adaboost which performs much better than its discrete version. In the system, uncorrelated features, color and texture are adopt to train two classifiers separately; the classifiers fused by voting generate confidence score for each pixel measuring its belonging to object or background in candidate regions; accumulated scores in each region are feed to particle filter for estimating object states; pixels with high scores augment the training set mutually and further classifiers are updated by co-training. The system is applied to vehicle and pedestrian tracking in real world scenarios and the experimental results show its robustness to large appearance variations and severe occlusions.

Family Photo Recognition via Multiple Instance Learning

Family photo recognition is an important task in social media analytics. Previous methods use singleton global features and conventional binary classifiers to distinguish family group photos from non-family ones. Different from them, we propose a novel family recognition approach with three dedicated local representations under Multiple Instance Learning framework, where geometry, kinship and semantic features are integrated to overcome issues in the previous work. Experimental results show that our method achieves the state-of-the-art result among global-feature models.

Robust road detection from a single image

Road detection from images is a challenging task in computer vision. Previous methods are not robust, because their features and classifiers cannot adapt to different circumstances. To overcome this problem, we propose to apply unsupervised feature learning for road detection. Specifically, we develop an improved encoding function and add a feature selection process to obtain robust and discriminative road features. Besides, a road segmentation algorithm is proposed to extract road regions from the learned feature maps, in which a tree structure is established to represent the hierarchical relations of various regions segmented by multiple thresholds, and a two-loop optimization is then employed to select the most stable regions as road areas. Experimental results on several challenging datasets justify the effectiveness of our method.

A genetics-motivated unsupervised model for tri-subject kinship verification

Given a childs and a couples facial photos, tri-subject kinship verification aims to determine the existence of blood relation between the child and the couple. Different from existing methods which model the kinship inheritance process among three persons in separate stages and only use simple features, this work establishes a simple model inspired by genetics to measure tri-subject kinship similarity in one step. Meanwhile, high-dimensional features are incorporated into this simple model to seek for better performance. Experiment results demonstrate the effectiveness of our approach.