Search

Accurate head pose estimation from 2D image data is an essential component of applications such as driver monitoring systems, virtual reality technology, and human-computer interaction. It enables a better determination of user engagement and attentiveness. The most accurate head pose estimators are based on Deep Neural Networks that are trained with the supervised approach and rely primarily on the accuracy of training data. The acquisition of real head pose data with a wide variation of yaw, pitch and roll is a challenging task. Publicly available head pose datasets have limitations with respect to size, resolution, annotation accuracy and diversity. In this work, a methodology is proposed to generate pixel-perfect synthetic 2D headshot images rendered from high-quality 3D synthetic facial models with accurate head pose annotations. A diverse range of variations in age, race, and gender are also provided. The resulting dataset includes more than 300k pairs of RGB images with corresponding head pose annotations. A wide range of variations in pose, illumination and background are included. The dataset is evaluated by training a state-of-the-art head pose estimation model and testing against the popular evaluation-dataset Biwi. The results show that training with purely synthetic data generated using the proposed methodology achieves close to state-of-the-art results on head pose estimation which are originally trained on real human facial datasets. As there is a domain gap between the synthetic images and real-world images in the feature space, initial experimental results fall short of the current state-of-the-art. To reduce the domain gap, a semi-supervised visual domain adaptation approach is proposed, which simultaneously trains with the labelled synthetic data and the unlabeled real data. When domain adaptation is applied, a significant improvement in model performance is achieved. Additionally, by applying a data fusion-based transfer learning approach, better results are achieved than previously published work on this topic.

Focusing on cross-dataset Automated Depression Recognition (ADR) by jointly exploring facial appearance and dynamics feature representations, we explore to propose a novel Latent Domain Adaptation Depression Recognition (latDADR) framework via discovering a discriminative domain-invariant subspace. In this subspace, the between-domain distribution discrepancy would be semantically minimized, and meanwhile the within-domain geometric structures would also be discriminatively preserved. In latDADR, we respectively optimize two target classifiers on dynamics features and appearance features as well as learn a source classifier on appearance features, and then encode certain shared components of the different domain classifiers as low-rank and sparse regularization terms. Moreover, the prediction results from two target classifiers are constrained to be consistent for better fusing the discriminative information from different feature representations. We specially use the l2,1 -norm based loss function for learning robust classifiers on different feature representations. Different from the state-of-the-arts, our method can borrow the discriminative information from another auxiliary domain for ADR, even if the target prior information is very scarce and the features of the source and target domains are partially different but overlapping. The proposed method is evaluated on three depression databases, and the experimental results demonstrate the superiorities and outstanding performance compared with several representative works.

Covariate shift assumption-based domain adaptation approaches usually utilize only one common transformation to align marginal distributions and make conditional distributions preserved. However, one common transformation may cause loss of useful information, such as variances and neighborhood relationship in both source and target domains. To address this problem, we propose a novel method called homologous component analysis (HCA) where we try to find two totally different but homologous transformations to align distributions with side information and make conditional distributions preserved. As it is hard to find a closed-form solution to the corresponding optimization problem, we solve them by means of the alternating direction minimizing method (ADMM) in the context of Stiefel manifolds. We also provide a generalization error bound for domain adaptation in the semi-supervised case, and two transformations can help to decrease this upper bound more than only one common transformation does. Extensive experiments on synthetic and real data show the effectiveness of the proposed method by comparing its classification accuracy with the state-of-the-art methods, and the numerical evidence on chordal distance and Frobenius distance shows that resulting optimal transformations are different. [ABSTRACT FROM AUTHOR]

Image processing has attracted increasing attention in recent researches to solve domain shift problem where machine learning algorithms are applied to sets of unseen images. Domain shift problem occurs when the training (source domain) and test (target domain) sets are collected in different environmental conditions but in related domains. In this way, the adaptation across data distributions of the source and target datasets are suggested as domain adaptation framework to overcome the performance degradation. In this paper, a novel domain adaptation method referred as joint local and statistical discriminant learning via feature alignment (LSA), is proposed to find a cross-domain subspace by matching cross-domain distribution shift and adapting the class structures of the local and statistical distributions across the source and target domains, during the dimensionality reduction. Specifically, LSA projects samples into an embedded subspace in which the distances across the samples from same class are minimized and the distances across samples from different classes are maximized, at each local and statistical area, during alignment of marginal and conditional distributions. Furthermore, the class densities of samples based on manifold structure in different classes are preserved to provide more separability across various classes. To evaluate the proposed method, comprehensive experiments have been conducted on benchmark cross-domain object and digit recognition datasets. Experimental results have verified the superiority of LSA with a large margin in average classification accuracy against several state-of-the-art distribution matching and discriminant learning methods of domain adaptation. Moreover, the results have demonstrated the effectiveness of our proposed representation learning. Our source code is available at https://github.com/jtahmores/LSA. [ABSTRACT FROM AUTHOR]

Purpose: Face detection is a needed component for the automatic analysis and assistance of human activities during surgical procedures. Efficient face detection algorithms can indeed help to detect and identify the persons present in the room and also be used to automatically anonymize the data. However, current algorithms trained on natural images do not generalize well to the operating room (OR) images. In this work, we provide a comparison of state-of-the-art face detectors on OR data and also present an approach to train a face detector for the OR by exploiting non-annotated OR images. Methods: We propose a comparison of six state-of-the-art face detectors on clinical data using multi-view OR faces, a dataset of OR images capturing real surgical activities. We then propose to use self-supervision, a domain adaptation method, for the task of face detection in the OR. The approach makes use of non-annotated images to fine-tune a state-of-the-art detector for the OR without using any human supervision. Results: The results show that the best model, namely the tiny face detector, yields an average precision of 0.556 at intersection over union of 0.5. Our self-supervised model using non-annotated clinical data outperforms this result by 9.2%. Conclusion: We present the first comparison of state-of-the-art face detectors on OR images and show that results can be significantly improved by using self-supervision on non-annotated data. [ABSTRACT FROM AUTHOR]

Information hiding and data encryption are used widely to protect data and information from anonymous access. In digital world, hiding and encrypting of the desired data into an image is a smart way to protect information with a low cost. In the digital images, steganalysis is a known method to distinguish between clean and stego images. Most of recent researches in this scope exploit feature reduction algorithms to improve the performance of correct detections. However, dimension reduction alone could not tackle the problem of steganalysis because the properties of stego images change during the steganalysis process. In this work, it is intended to propose an Image Steganalysis using visual Domain Adaptation (ISDA), which this steganalysis target images to distinguish across stego and clean images. ISDA is a dimensionality reduction approach that considers the image drifts during the steganography process in the steganalysis of target images. Moreover, ISDA employs domain invariant clustering in an embedded representation to cluster clean and stego images in the reduced subspace. The results on benchmark datasets demonstrate that ISDA thoroughly outperforms all of the state of the art methods on validation parameters, accuracy of detection and time complexity

Information hiding and data encryption are used widely to protect data and information from anonymous access. In digital world, hiding and encrypting of the desired data into an image is a smart way to protect information with a low cost. In the digital images, steganalysis is a known method to distinguish between clean and stego images. Most of recent researches in this scope exploit feature reduction algorithms to improve the performance of correct detections. However, dimension reduction alone could not tackle the problem of steganalysis because the properties of stego images change during the steganalysis process. In this work, it is intended to propose an Image Steganalysis using visual Domain Adaptation (ISDA), which this steganalysis target images to distinguish across stego and clean images. ISDA is a dimensionality reduction approach that considers the image drifts during the steganography process in the steganalysis of target images. Moreover, ISDA employs domain invariant clustering in an embedded representation to cluster clean and stego images in the reduced subspace. The results on benchmark datasets demonstrate that ISDA thoroughly outperforms all of the state of the art methods on validation parameters, accuracy of detection and time complexity. [ABSTRACT FROM AUTHOR]

In real-world computer vision applications, many intrinsic and extrinsic variations can cause a significant domain shift. Although deep convolutional models have provided us with better domain-invariant features, existing mechanisms to adapt convolutional activations are still limited. Notice that convolutional activations are intrinsically represented as tensors, in this paper we develop a two-dimensional subspace alignment (2DSA) approach based on 2D principal component analysis (PCA) to better adapt convolutional activations. Extensive experiments demonstrate the advantages of 2DSA over its counterpart SA in both effectiveness and efficiency. In particular, when trying to explain why 2DSA works well, we find that the best classification performance has low correlation with the global domain discrepancy measure. In an effort to find a better way to compare domains, we introduce within- and between-class domain divergence measures to characterize the class-level differences. The proposed measures somewhat shed light on what a good alignment might be for classification. Furthermore, we also demonstrate a novel domain adaptation application in agriculture and create a dataset for the problem. [ABSTRACT FROM AUTHOR]

Visual domain adaptation has attracted much attention and has made great achievement in recent years. It deals with the problem of distribution divergence between source and target domains. Current methods mostly focus on transforming images from different domains into a common space to minimize the distribution divergence. However, there are many irrelevant source samples for target domain even after the transformation. In order to eliminate the irrelevant samples, we develop a sample selection algorithm using sparse coding theory. We do the sample selection in a common subspace of source and target data to find as many as relevant source samples. In the common subspace, data characteristics are preserved by using graph regularization. Therefore, we can select the most relevant samples for our target image classification task. Moreover, in order to build a discriminative classifier for the target domain, we use not only the common part of source and target domains learned in the common subspace but also the specific part of target domain. The algorithm can be extended to handle samples from multiple source domains. Experimental results show that our visual domain adaptation method on the image classification tasks can be very effective for the state-of-the-art datasets. [ABSTRACT FROM AUTHOR]