arXiv subjects

Jiasong Sun

Publications and source records attributed to Jiasong Sun.

12 recordsLinked to original sources

arXiv2025

Doubling the field of view and eliminating the replica overlap problem in common-path shearing quantitative phase imaging

Quantitative phase imaging (QPI) enables label-free, high-contrast visualization of transparent specimens, but its common implementation in off-axis digital holographic microscopy (DHM) requires a separate reference beam, which increases system complexity and sensitivity to noise and vibrations. Common-path shearing DHMs overcome these drawbacks by eliminating the reference arm, yet they suffer from sheared object beam (replica) overlap, as both interfering sheared beams traverse the sample and generate superimposed phase images. This limits their use to sparse objects only. Here we introduce R2D-QPI, a method that numerically decouples object and replica fields of view through controlled shear scanning. The method analytically separates overlapped phase images and effectively doubles the imaged area, requiring only two measurements. We experimentally validate the approach on a phase resolution test target, yeast cells, and human thyroid tissue slices, demonstrating accurate reconstruction even in highly confluent samples with strong object-replica overlap. The results establish R2D-QPI as a robust and versatile solution for common-path QPI, enabling wide-field, label-free phase imaging with minimal data acquisition and strong potential for applications in biological and medical microscopy.

physics.optics↗

arXiv2025

Differentiable Imaging Meets Adaptive Neural Dropout: An Advancing Method for Transparent Object Tomography

Label-free tomographic microscopy offers a compelling means to visualize three-dimensional (3D) refractive index (RI) distributions from two-dimensional (2D) intensity measurements. However, limited forward-model accuracy and the ill-posed nature of the inverse problem hamper artifact-free reconstructions. Meanwhile, artificial neural networks excel at modeling nonlinearities. Here, we employ a Differentiable Imaging framework that represents the 3D sample as a multi-layer neural network embedding physical constraints of light propagation. Building on this formulation, we propose a physics-guided Adaptive Dropout Neural Network (ADNN) for optical diffraction tomography (ODT), focusing on network topology and voxel-wise RI fidelity rather than solely on input-output mappings. By exploiting prior knowledge of the sample's RI, the ADNN adaptively drops and reactivates neurons, enhancing reconstruction accuracy and stability. We validate this method with extensive simulations and experiments on weakly and multiple-scattering samples under different imaging setups. The ADNN significantly improves quantitative 3D RI reconstructions, providing superior optical-sectioning and effectively suppressing artifacts. Experimental results show that the ADNN reduces the Mean Absolute Error (MAE) by a factor of 3 to 5 and increases the Structural Similarity Index Metric (SSIM) by about 4 to 30 times compared to the state-of-the-art approach.

physics.optics↗

arXiv2020

On a universal solution to the transport-of-intensity equation

Transport-of-intensity equation (TIE) is one of the most well-known approaches for phase retrieval and quantitative phase imaging. It directly recovers the quantitative phase distribution of an optical field by through-focus intensity measurements in a noninterferometic, deterministic manner. Nevertheless, the accuracy and validity of state-of-the-art TIE solvers depend on restrictive preknowledge or assumptions, including appropriate boundary conditions, a well-defined closed region, and quasi-uniform in-focus intensity distribution, which, however, cannot be strictly satisfied simultaneously under practical experimental conditions. In this Letter, we propose a universal solution to TIE with the advantages of high accuracy, convergence guarantee, applicability to arbitrarily-shaped regions, and simplified implementation and computation. With the "maximum intensity assumption", we firstly simplified TIE as a standard Possion equation to get an initial guess of the solution. Then the initial solution is further refined iteratively by solving the same Possion equation, and thus, the instability associated with the division by zero/small intensity values and large intensity variations can be effectively bypassed. Simulations and experiments with arbitrary phase, arbitrary aperture shapes, and nonuniform intensity distributions verify the effectiveness and universality of the proposed method.

eess.IV↗

arXiv2019

Utterance-level Permutation Invariant Training with Latency-controlled BLSTM for Single-channel Multi-talker Speech Separation

Utterance-level permutation invariant training (uPIT) has achieved promising progress on single-channel multi-talker speech separation task. Long short-term memory (LSTM) and bidirectional LSTM (BLSTM) are widely used as the separation networks of uPIT, i.e. uPIT-LSTM and uPIT-BLSTM. uPIT-LSTM has lower latency but worse performance, while uPIT-BLSTM has better performance but higher latency. In this paper, we propose using latency-controlled BLSTM (LC-BLSTM) during inference to fulfill low-latency and good-performance speech separation. To find a better training strategy for BLSTM-based separation network, chunk-level PIT (cPIT) and uPIT are compared. The experimental results show that uPIT outperforms cPIT when LC-BLSTM is used during inference. It is also found that the inter-chunk speaker tracing (ST) can further improve the separation performance of uPIT-LC-BLSTM. Evaluated on the WSJ0 two-talker mixed-speech separation task, the absolute gap of signal-to-distortion ratio (SDR) between uPIT-BLSTM and uPIT-LC-BLSTM is reduced to within 0.7 dB.

cs.SD↗

arXiv2019

Resolution analysis in a lens-free on-chip digital holographic microscope

Lens-free on-chip digital holographic microscopy (LFOCDHM) is a modern imaging technique whereby the sample is placed directly onto or very close to the digital sensor, and illuminated by a partially coherent source located far above it. The scattered object wave interferes with the reference (unscattered) wave at the plane where a digital sensor is situated, producing a digital hologram that can be processed in several ways to extract and numerically reconstruct an in-focus image using the back propagation algorithm. Without requiring any lenses and other intermediate optical components, the LFOCDHM has unique advantages of offering a large effective numerical aperture (NA) close to unity across the native wide field-of-view (FOV) of the imaging sensor in a cost-effective and compact design. However, unlike conventional coherent diffraction limited imaging systems, where the limiting aperture is used to define the system performance, typical lens-free microscopes only produce compromised imaging resolution that far below the ideal coherent diffraction limit. At least five major factors may contribute to this limitation, namely, the sample-to-sensor distance, spatial and temporal coherence of the illumination, finite size of the equally spaced sensor pixels, and finite extent of the image sub-FOV used for the reconstruction, which have not been systematically and rigorously explored until now. In this work, we derive five transfer function models that account for all these physical effects and interactions of these models on the imaging resolution of LFOCDHM. We also examine how our theoretical models can be utilized to optimize the optical design or predict the theoretical resolution limit of a given LFOCDHM system. We present a series of simulations and experiments to confirm the validity of our theoretical models.

physics.optics↗

arXiv2019

Wide-field high-resolution 3D microscopy with Fourier ptychographic diffraction tomography

We report a computational 3D microscopy technique, termed Fourier ptychographic diffraction tomography (FPDT), that iteratively stitches together numerous variably illuminated, low-resolution images acquired with a low-numerical aperture (NA) objective in 3D Fourier space to create a wide field-of-view (FOV), high-resolution, depth-resolved complex refractive index (RI) image across large volumes. Unlike conventional optical diffraction tomography (ODT) approaches that rely on controlled bright-field illumination, holographic phase measurement, and high-NA objective detection, FPDT employs tomographic RI reconstruction from low-NA intensity-only measurements. In addition, FPDT incorporates high-angle dark-field illuminations beyond the NA of the objective, significantly expanding the accessible object frequency. With FPDT, we present the highest-throughput ODT results with 390nm lateral resolution and 899nm axial resolution across a 10X FOV of 1.77mm2 and a depth of focus of ~20μm. Billion-voxel 3D tomographic imaging results of biological samples establish FPDT as a powerful non-invasive and label-free tool for high-throughput 3D microscopy applications.

physics.optics↗

arXiv2019

Optimal illumination scheme for isotropic quantitative differential phase contrast microscopy

Differential phase contrast microscopy (DPC) provides high-resolution quantitative phase distribution of thin transparent samples under multi-axis asymmetric illuminations. Typically, illumination in DPC microscopic systems is designed with 2-axis half-circle amplitude patterns, which, however, result in a non-isotropic phase contrast transfer function (PTF). Efforts have been made to achieve isotropic DPC by replacing the conventional half-circle illumination aperture with radially asymmetric patterns with 3-axis illumination or gradient amplitude patterns with 2-axis illumination. Nevertheless, these illumination apertures were empirically designed based on empirical criteria related to the shape of the PTF, leaving the underlying theoretical mechanisms unexplored. Furthermore, the frequency responses of the PTFs under these engineered illuminations have not been fully optimized, leading to suboptimal phase contrast and signal-to-noise ratio (SNR) for phase reconstruction. In this Letter, we provide a rigorous theoretical analysis about the necessary and sufficient conditions for DPC to achieve perfectly isotropic PTF. In addition, we derive the optimal illumination scheme to maximize the frequency response for both low and high frequencies (from 0 to 2NAobj), and meanwhile achieve perfectly isotropic PTF with only 2-axis intensity measurements. We present the derivation, implementation, simulation and experimental results demonstrating the superiority of our method over state-of-the-arts in both phase reconstruction accuracy and noise-robustness.

physics.optics↗

arXiv2018

Angular Softmax Loss for End-to-end Speaker Verification

End-to-end speaker verification systems have received increasing interests. The traditional i-vector approach trains a generative model (basically a factor-analysis model) to extract i-vectors as speaker embeddings. In contrast, the end-to-end approach directly trains a discriminative model (often a neural network) to learn discriminative speaker embeddings; a crucial component is the training criterion. In this paper, we use angular softmax (A-softmax), which is originally proposed for face verification, as the loss function for feature learning in end-to-end speaker verification. By introducing margins between classes into softmax loss, A-softmax can learn more discriminative features than softmax loss and triplet loss, and at the same time, is easy and stable for usage. We make two contributions in this work. 1) We introduce A-softmax loss into end-to-end speaker verification and achieve significant EER reductions. 2) We find that the combination of using A-softmax in training the front-end and using PLDA in the back-end scoring further boosts the performance of end-to-end systems under short utterance condition (short in both enrollment and test). Experiments are conducted on part of $Fisher$ dataset and demonstrate the improvements of using A-softmax.

eess.AS↗

arXiv2018

Optimal illumination pattern for transport-of-intensity quantitative phase microscopy

The transport-of-intensity equation (TIE) is a well-established non-interferometric phase retrieval approach, which enables quantitative phase imaging (QPI) of transparent sample simply by measuring the intensities at multiple axially displaced planes. Nevertheless, it still suffers from two fundamentally limitations. First, it is quite susceptible to low-frequency errors (such as \cloudy" artifacts), which results from the poor contrast of the phase transfer function (PTF) near the zero frequency. Second, the reconstructed phase tends to blur under spatially low-coherent illumination, especially when the defocus distance is beyond the near Fresnel region. Recent studies have shown that the shape of the illumination aperture has a significant impact on the resolution and phase reconstruction quality, and by simply replacing the conventional circular illumination aperture with an annular one, these two limitations can be addressed, or at least significantly alleviated. However, the annular aperture was previously empirically designed based on intuitive criteria related to the shape of PTF, which does not guarantee optimality. In this work, we optimize the illumination pattern to maximize TIE's performance based on a combined quantitative criterion for evaluating the \goodness" of an aperture. In order to make the size of the solution search space tractable, we restrict our attention to binary coded axis-symmetric illumination patterns only, which are easier to implement and can generate isotropic TIE PTFs. We test the obtained optimal illumination by imaging both a phase resolution target and HeLa cells based on a small-pitch LED array, suggesting superior performance over other suboptimal patterns in terms of both signal-to-noise ratio (SNR) and spatial resolution.

eess.IV↗

arXiv2017

An Improved Residual LSTM Architecture for Acoustic Modeling

Long Short-Term Memory (LSTM) is the primary recurrent neural networks architecture for acoustic modeling in automatic speech recognition systems. Residual learning is an efficient method to help neural networks converge easier and faster. In this paper, we propose several types of residual LSTM methods for our acoustic modeling. Our experiments indicate that, compared with classic LSTM, our architecture shows more than 8% relative reduction in Phone Error Rate (PER) on TIMIT tasks. At the same time, our residual fast LSTM approach shows 4% relative reduction in PER on the same task. Besides, we find that all this architecture could have good results on THCHS-30, Librispeech and Switchboard corpora.

cs.CL↗

arXiv2017

Adaptive pixel-super-resolved lensfree holography for wide-field on-chip microscopy

High-resolution wide field-of-view (FOV) microscopic imaging plays an essential role in various fields of biomedicine, engineering, and physical sciences. As an alternative to conventional lens-based scanning techniques, lensfree holography provides a new way to effectively bypass the intrinsical trade-off between the spatial resolution and FOV of conventional microscopes. Unfortunately, due to the limited sensor pixel-size, unpredictable disturbance during image acquisition, and sub-optimum solution to the phase retrieval problem, typical lensfree microscopes only produce compromised imaging quality in terms of lateral resolution and signal-to-noise ratio (SNR). Here, we propose an adaptive pixel-super-resolved lensfree imaging (APLI) method which can solve, or at least partially alleviate these limitations. Our approach addresses the pixel aliasing problem by Z-scanning only, without resorting to subpixel shifting or beam-angle manipulation. Automatic positional error correction algorithm and adaptive relaxation strategy are introduced to enhance the robustness and SNR of reconstruction significantly. Based on APLI, we perform full-FOV reconstruction of a USAF resolution target ($\sim$29.85 $m{m^2}$) and achieve half-pitch lateral resolution of 770 $nm$, surpassing 2.17 times of the theoretical Nyquist-Shannon sampling resolution limit imposed by the sensor pixel-size (1.67 $μm$). Full-FOV imaging result of a typical dicot root is also provided to demonstrate its promising potential applications in biologic imaging.

physics.optics↗

arXiv2017

High-resolution transport-of-intensity quantitative phase microscopy with annular illumination

For quantitative phase imaging (QPI) based on transport-of-intensity equation (TIE), partially coherent illumination provides speckle-free imaging, compatibility with brightfield microscopy, and transverse resolution beyond coherent diffraction limit. Unfortunately, in a conventional microscope with circular illumination aperture, partial coherence tends to diminish the phase contrast, exacerbating the inherent noise-to-resolution tradeoff in TIE imaging, resulting in strong low-frequency artifacts and compromised imaging resolution. Here, we demonstrate how these issues can be effectively addressed by replacing the conventional circular illumination aperture with an annular one. The matched annular illumination not only strongly boosts the phase contrast for low spatial frequencies, but significantly improves the practical imaging resolution to near the incoherent diffraction limit. By incorporating high-numerical aperture (NA) illumination as well as high-NA objective, it is shown, for the first time, that TIE phase imaging can achieve a transverse resolution up to 208 nm, corresponding to an effective NA of 2.66. Time-lapse imaging of in vitro Hela cells revealing cellular morphology and subcellular dynamics during cells mitosis and apoptosis is exemplified. Given its capability for high-resolution QPI as well as the compatibility with widely available brightfield microscopy hardware, the proposed approach is expected to be adopted by the wider biology and medicine community.

physics.optics↗