Effective training of the deep neural networks requires much data to avoid underdetermined and poor generalization. Data Augmentation alleviates this by using existing more effectively. However standard augmentation produces only limited plausible alternative for example, flipping, distorting, adding noise to, cropping a patch from original samples. In paper, we introduce adversarial autoencoder (AAE) impose feature representations with uniform distribution apply linear interpolation on...

We consider the problem of comparing similarity image sets with variable-quantity, quality and un-ordered heterogeneous images. use feature restructuring to exploit correlations both inner&inter-set Specifically, residual self-attention can effectively restructure features using other within a set emphasize discriminative images eliminate redundancy. Then, sparse/collaborative learning-based dependency-guided representation scheme reconstructs probe conditional gallery in order adaptively...

Unsupervised domain adaptation (UDA) aims to transfer the knowledge on a labeled source distribution perform well an unlabeled target domain. Recently, deep self-training involves iterative process of predicting and then taking confident predictions as hard pseudo-labels for retraining. However, are usually unreliable, easily leading deviated solutions with propagated errors. In this paper, we resort energy-based model constrain training sample energy function minimization objective. It can...

Various representation-based methods have been developed and shown great potential for pattern classification. To further improve their discriminability, we propose a Bi-level optimization framework in terms of both low-dimensional projection collaborative representation. Specifically, during the phase, try to minimize intra-class similarity inter-class dissimilarity, while representation our goal is achieve lowest correlation results. Solving this joint mutually reinforces aspects feature...

An increasing number of heavy machinery and vehicles have come into service, giving rise to a significant concern over protecting these high-security systems from misuse. Conventionally, authentication performed merely at the initial login may not be sufficient for detecting intruders throughout operating session. To address this critical security flaw, line-scan continuous hand system with appearance an rod is proposed. Given that occupied period, it can possible solution unobtrusively...

This paper targets to explore the inter-subject variations eliminated facial expression representation in compressed video domain. Most of previous methods process RGB images a sequence, while off-the-shelf and valuable expression-related muscle movement already embedded compression format. In up two orders magnitude domain, we can explicitly infer from residual frames possible extract identity factors I frame with pre-trained face recognition network. By enforcing marginal independent them,...

In order to minimize moiré patterns in autostereoscopic parallax displays, the optical component, which is used for forming viewing zones, analyzed with varying period and slant angle. First, horizontal-parallax displays (HPAD) can be approximated as superposition of three corresponding binary gratings. Referring unification indicial equation method Fourier analysis, a singular state two stable moiré-free states are obtained according equivalent grating LCD special radial grating. Two valid...

Deep learning is usually data starved, and the unsupervised domain adaptation (UDA) developed to introduce knowledge in labeled source unlabeled target domain. Recently, deep self-training presents a powerful means for UDA, involving an iterative process of predicting then taking confident predictions as hard pseudo-labels retraining. However, are unreliable, thus easily leading deviated solutions with propagated errors. In this paper, we resort energy-based model constrain training sample...

We propose a parameter design of the parallax barrier (PB) based on color moiré patterns in autostereoscopic displays. First, display device and PB are approximated as two corresponding binary gratings. In order to obtain different predominant Fourier low-frequency terms, superposition equivalent grating for special radial is analyzed, referring indicial equation method theory. Moreover, transition regions considered where vary gently. Finally, appropriate can be obtained. The validity...

We propose an interpretation of moiré phenomenon in the image domain. The is basically based on analysis waveform line families. period, angle, and intensity profile fringes can be obtained directly domain according to this interpretation. Moreover, pseudo-moiré interpreted visually with consideration illusional contrast human visual system. interpretation, which consistent Fourier theory when two superposed gratings are periodic, involves only shows remarkable simplicity, just like indicial...

In this Letter, we present an electrically tunable holographic waveguide display (HWD) based on two slanted polymer dispersed liquid crystal (HPDLC) gratings. Experimental results show that a see-through effect is obtained in the HWD both light from and ambient can be clearly seen simultaneously. By applying external electric field, output intensity of modulated, which attributed to field-induced rotation molecules HPDLC We also performance enables adapt different conditions. This study...

In this Letter, we present a display system based on curved screen and parallax barrier, which provides stereo images with horizontal field of view 360° without wearing any eyewear, to achieve an immersive autostereoscopic effect. The principle characteristics are studied theoretically in detail. Three consecutive pixels barrier form unit, can generate separate viewing zones for the left right eyes, respectively. Simulation experimental results show that non-crosstalk effect be obtained...

Panoramic stereo images, captured by distributed devices then mosaicking, are competent contents for virtual reality applications. Mosaicking raw images with different perspectives into satisfying final results is still not efficient enough, even if state-of-the-art algorithms employed. For improving this efficiency in optical methods, we delve the potential of capturing system. Two parallax factors, peak and deviation parallaxes, proposed to assess mosaicking capability. By controlling...

In order to realize autostereoscopic 3D projecting display with higher brightness and larger size, a technique of using parallax barriers lenticular sheets is proposed.The parameter design investigated. First, projection distance projector array confirmed according eye's resolution,then the strips' character analysed by zemax in four-view situation,thickness parameters strips are confirmed. Finally, width slit pitch viewing etc sheets. Data analysis indicate that our image increased three...

Structured light-based 3D sensing techniques has been an important means for digitization of real objects. However, to obtain a complete model target, multiple scans are demanded and that makes the operation time-consuming. In this paper, efficient scanning system is introduced, which composed two structured light sensors. The sensors placed in opposite directions, target between them. With single scan operation, point clouds can be obtained aligned by proposed calibration parameters. For...

Unsupervised domain adaptation (UDA) aims to transfer the knowledge on a labeled source distribution perform well an unlabeled target domain. Recently, deep self-training involves iterative process of predicting and then taking confident predictions as hard pseudo-labels for retraining. However, are usually unreliable, easily leading deviated solutions with propagated errors. In this paper, we resort energy-based model constrain training sample energy function minimization objective. It can...

This paper targets to explore the inter-subject variations eliminated facial expression representation in compressed video domain. Most of previous methods process RGB images a sequence, while off-the-shelf and valuable expression-related muscle movement already embedded compression format. In up two orders magnitude domain, we can explicitly infer from residual frames possible extract identity factors I frame with pre-trained face recognition network. By enforcing marginal independent them,...