Phase retrieval from single-shot square wave fringe based on image denoising using deep learning.

Fringe-structured light measurement technology has garnered significant attention in recent years. To enhance measurement speed while maintaining a certain level of accuracy using binary fringe, this paper proposes a phase retrieval method with single-frame binary square wave fringe. The proposed method utilizes image denoising through deep learning to extract the phase, enabling the use of a trained image denoiser as a low-pass filter, which adaptively replaces the manual selection of the appropriate band-pass filter. The results demonstrate that this method achieves higher reconstruction accuracy than the traditional single-frame algorithm while preserving more object details.

High accuracy, compact 3D face imaging method based on translational and online-switchable phase-shifting fringe projection.

In this paper, a compact, cost-effective, and fast translational online-switchable phase-shifting fringe (TOPF) projector is designed and fabricated for high accuracy three-dimensional (3D) face imaging. Compared with the conventional mechanical projectors, the main difference is that it utilizes a translational approach instead of a rotational one to achieve a better balance in terms of size, speed, accuracy, and cost. To mitigate the inconsistency of the motor's step size and ensure the stability of phase-shifting, an optical encoder-based feedback control mechanism is employed. Additionally, to address the random phase shift errors induced by mechanical motion, a fast, generalized phase-shifting algorithm with unknown phase shifts (uPSAs) that can calculate arbitrary phase shifts is proposed. Finally, a 3D imaging system consisting of the TOPF projector and two cameras is constructed for experimental validation. The feasibility, effectiveness, and precision of our proposed method are substantiated through the reconstruction of a static facial model and a dynamic real face.

3D face recognition algorithm based on deep Laplacian pyramid under the normalization of epidemic control.

Under the normalization of epidemic control in COVID-19, it is essential to realize fast and high-precision face recognition without feeling for epidemic prevention and control. This paper proposes an innovative Laplacian pyramid algorithm for deep 3D face recognition, which can be used in public. Through multi-mode fusion, dense 3D alignment and multi-scale residual fusion are ensured. Firstly, the 2D to 3D structure representation method is used to fully correlate the information of crucial points, and dense alignment modeling is carried out. Then, based on the 3D critical point model, a five-layer Laplacian depth network is constructed. High-precision recognition can be achieved by multi-scale and multi-modal mapping and reconstruction of 3D face depth images. Finally, in the training process, the multi-scale residual weight is embedded into the loss function to improve the network's performance. In addition, to achieve high real-time performance, our network is designed in an end-to-end cascade. While ensuring the accuracy of identification, it guarantees personnel screening under the normalization of epidemic control. This ensures fast and high-precision face recognition and establishes a 3D face database. This method is adaptable and robust in harsh, low light, and noise environments. Moreover, it can complete face reconstruction and recognize various skin colors and postures.

Theoretical analysis and experimental investigation of the Floyd-Steinberg-based fringe binary method with offset compensation for accurate 3D measurement.

Digital fringe projection (DFP) with defocused binary fringe patterns has the ability to overcome the projector nonlinearity and achieve a high-speed 3D measurement. The Floyd-Steinberg (FS) dithering technique is one of the most commonly adopted binary fringe coding methods due to its relatively high measurement accuracy. Nevertheless, we found that the FS binary fringe would cause a fixed error in the recovered phase, which is proven to be invariable for various defocusing levels and various phase-shift steps according to the analysis of the phase error based on noise model of phase-shifting profilometry. It means that FS binary fringe would have a certain offset in space, compared with standard sinusoidal fringe, which is verified to be essentially constant for different fringe pitches through simulation and experiment. This offset would distort the 3D geometry of the tested target for monocular systems relying on triangulation, which needs to be compensated to improve 3D measurement accuracy. Experiments are presented to demonstrate the enhanced 3D result after compensation.

A selective kernel-based cycle-consistent generative adversarial network for unpaired low-dose CT denoising.

Low-dose computed tomography (LDCT) denoising is an indispensable procedure in the medical imaging field, which not only improves image quality, but can mitigate the potential hazard to patients caused by routine doses. Despite the improvement in performance of the cycle-consistent generative adversarial network (CycleGAN) due to the well-paired CT images shortage, there is still a need to further reduce image noise while retaining detailed features. Inspired by the residual encoder-decoder convolutional neural network (RED-CNN) and U-Net, we propose a novel unsupervised model using CycleGAN for LDCT imaging, which injects a two-sided network into selective kernel networks (SK-NET) to adaptively select features, and uses the patchGAN discriminator to generate CT images with more detail maintenance, aided by added perceptual loss. Based on patch-based training, the experimental results demonstrated that the proposed SKFCycleGAN outperforms competing methods in both a clinical dataset and the Mayo dataset. The main advantages of our method lie in noise suppression and edge preservation.

Optical random phase dropout in a diffractive deep neural network.

Unitary learning is a backpropagation (BP) method that serves to update unitary weights in fully connected deep complex-valued neural networks, meeting a prior unitary in an active modulation diffractive deep neural network. However, the square matrix characteristic of unitary weights in each layer results in its learning belonging to a small-sample training, which produces an almost useless network that has a fairly poor generalization capability. To alleviate such a serious over-fitting problem, in this Letter, optical random phase dropout is formulated and designed. The equivalence between unitary forward and diffractive networks deduces a synthetic mask that is seamlessly compounded with a computational modulation and a random sampling comb called dropout. The dropout is filled with random phases in its zero positions that satisfy the Bernoulli distribution, which could slightly deflect parts of transmitted optical rays in each output end to generate statistical inference networks. The enhancement of generalization benefits from the fact that massively parallel full connection with different optical links is involved in the training. The random phase comb introduced into unitary BP is in the form of conjugation, which indicates the significance of optical BP.

3-D face registration solution with speckle encoding based spatial-temporal logical correlation algorithm.

3-D information acquisition (registration) of whole face plays a significant role in 3-D human face recognition application. In this paper, we develop a prototype of 3-D system consisting of two binocular measurement units that allows a full 3-D reconstruction by utilizing the advantages of a novel correlation algorithm. In this system, we use optical modulation to produce temporally and spatially varying high-density binary speckle patterns to encode the tested face, then propose a spatial-temporal logical correlation (STLC) stereo matching algorithm to fast determine the accurate disparity with a coarse and refined strategy. Finally the 3-D information of whole face from left- and right ear (~180°) can be obtainable by fusing the data from two measurement units. Comparative researches are performed to test a plastic model and a real human face by simulating real application situations. The results verify the feasibility and good performances of our computational frameworks and experimental configuration in terms of accuracy and time cost, which show a good application prospect in our future 3-D human face recognition research.

Sparsity-based continuous wave terahertz lens-free on-chip holography with sub-wavelength resolution.

We demonstrate terahertz (THz) lens-free in-line holography on a chip in order to achieve 40 µm spatial resolution corresponding to ~0.7λ with a numerical aperture of ~0.87. We believe that this is the first time that sub-wavelength resolution in THz holography and the 40 µm resolution were both far better than what was already reported. The setup is based on a self-developed high-power continuous wave THz laser at 5.24 THz (λ = 57.25 µm) and a high-resolution microbolometer detector array (640 × 512 pixels) with a pitch of 17 µm. This on-chip in-line holography, however, suffers from the twin-image artifacts which obfuscate the reconstruction. To address this problem, we propose an iterative optimization framework, where the conventional object constraint and the L1 sparsity constraint can be combined to efficiently reconstruct the complex amplitude distribution of the sample. Note that the proposed framework and the sparsity-based algorithm can be applied to holography in other wavebands without limitation of wavelength. We demonstrate the success of this sparsity-based on-chip holography by imaging biological samples (i.e., a dragonfly wing and a bauhinia leaf).

Experimental study of temporal-spatial binary pattern projection for 3D shape acquisition.

Three-dimensional (3D) acquisition of an object with modest accuracy and speed is of particular concern in practice. The performance of digital sinusoidal fringe pattern projection using an off-the-shelf digital video projector is generally discounted by the nonlinearity and low switch rate. In this paper, a binary encoding method to encode one computer-generated standard sinusoidal fringe pattern is presented for circumventing such deficiencies. In previous work [Opt. Eng.54, 054108 (2015)OPEGAR0091-328610.1117/1.OE.54.5.054108], we have developed a 3D system based on this encoding tactic and showed its prospective application. Here, we first build a physical model to explain the mechanism of how to generate good sinusoidality. The phase accuracy with respect to the conventional spatial binary encoding method and sinusoidal fringe pattern is also comparatively evaluated through simulation and experiments. We also adopt two phase-height mapping relationships to experimentally compare the measurement accuracy among them. The results indicate that the proposed binary encoding strategy has a comparable performance to that of sinusoidal fringe pattern projection and enjoys advantages over the spatial binary method under the same conditions.

Accurate and fast 3D surface measurement with temporal-spatial binary encoding structured illumination.

Balancing the accuracy and speed for 3D surface measurement of object is crucial in many important applications. Binary encoding pattern utilizing the high-speed image switching rate of digital mirror device (DMD)-based projector could be used as the candidate for fast even high-speed 3D measurement, but current most schemes only enable the measurement speed, which limit their application scopes. In this paper, we present a binary encoding method and develop an experimental system aiming to solve such a situation. Our approach encodes one computer-generated standard 8 bit sinusoidal fringe pattern into multiple binary patterns (sequence) with designed temporal-spatial binary encoding tactics. The binary pattern sequence is then high-speed and in-focus projected onto the surface of tested object, and then captured by means of temporal-integration imaging to form one sinusoidal fringe image. Further the combination of phase-shifting technique and temporal phase unwrapping algorithm leads to fast and accurate 3D measurement. The systematic accuracy better than 0.08mm is achievable. The measurement results with mask and palm are given to confirm the feasibility.

Modulation measuring profilometry with auto-synchronous phase shifting and vertical scanning.

To determine the shape of a complex object with vertical measurement mode and higher accuracy, a novel modulation measuring profilometry realizing auto-synchronous phase shifting and vertical scanning is proposed. Coaxial optical system for projection and observation instead of triangulation system is adopted to avoid shadow and occlusion. In the projecting system, sinusoidal grating is perpendicular to optical axis. For moving the grating along a direction at a certain angle to optical axis, 1D precision translation platform is applied to achieve purposes of both phase-shifting and vertical scanning. A series of fringe patterns with different modulation variations are captured by a CCD camera while scanning. The profile of the tested object can be reconstructed by the relationship between the height values and the modulation distributions. Unlike the previous method based on Fourier transform for 2D fringe pattern, the modulation maps are calculated from the intensity curve formed by the points with definite pixel coordinates in the captured fringe patterns. The paper gives the principle of the proposed method, the set-up of measurement system and the method for system calibration. Computer simulation and experiment results proved its feasibility.

k-NS: a classifier by the distance to the nearest subspace.

To improve the classification performance of k-NN, this paper presents a classifier, called k -NS, based on the Euclidian distances from a query sample to the nearest subspaces. Each nearest subspace is spanned by k nearest samples of a same class. A simple discriminant is derived to calculate the distances due to the geometric meaning of the Grammian, and the calculation stability of the discriminant is guaranteed by embedding Tikhonov regularization. The proposed classifier, k-NS, categorizes a query sample into the class whose corresponding subspace is proximal. Because the Grammian only involves inner products, the classifier is naturally extended into the high-dimensional feature space induced by kernel functions. The experimental results on 13 publicly available benchmark datasets show that k-NS is quite promising compared to several other classifiers founded on nearest neighbors in terms of training and test accuracy and efficiency.

On the almost periodic solution of cellular neural networks with distributed delays.

By exponential dichotomy about differential equations, a formal almost periodic solution (APS) of a class of cellular neural networks (CNNs) with distributed delays is obtained. Then, within different normed spaces, several sufficient conditions guaranteeing the existence and uniqueness of an APS are proposed using two fixed-point theorems. Based on the continuity property and some inequality techniques, two theorems insuring the global stability of the unique APS are given. Comparing with known literatures, all conclusions are drawn with slacker restrictions, e.g., do not require the integral of the kernel function determining the distributed delays from zero to positive infinity to be one, and the activation functions to be bounded, etc.; besides, all criteria are obtained by different ways. Finally, two illustrative examples show the validity and that all criteria are easy to check and apply.

A functional neural network computing some eigenvalues and eigenvectors of a special real matrix.

How to quickly compute eigenvalues and eigenvectors of a matrix, especially, a general real matrix, is significant in engineering. Since neural network runs in asynchronous and concurrent manner, and can achieve high rapidity, this paper designs a concise functional neural network (FNN) to extract some eigenvalues and eigenvectors of a special real matrix. After equivalent transforming the FNN into a complex differential equation and obtaining the analytic solution, the convergence properties of the FNN are analyzed. If the eigenvalue whose imaginary part is nonzero and the largest of all eigenvalues is unique, the FNN will converge to the eigenvector corresponding to this special eigenvalue with general nonzero initial vector. If all eigenvalues are real numbers or there are more than one eigenvalue whose imaginary part equals the largest, the FNN will converge to zero point or fall into a cycle procedure. Comparing with other neural networks designed for the same domain, the restriction to matrix is very slack. At last, three examples are employed to illustrate the performance of the FNN.