A5038566732- Advanced Image and Video Retrieval Techniques
- Cell Image Analysis Techniques
- Visual Attention and Saliency Detection
- Video Surveillance and Tracking Methods
- Advanced Vision and Imaging
- Medical Image Segmentation Techniques
- Reproductive Biology and Fertility
- Domain Adaptation and Few-Shot Learning
- Single-cell and spatial transcriptomics
- Video Analysis and Summarization
- AI in cancer detection
- Metabolomics and Mass Spectrometry Studies
- Advanced Neural Network Applications
- Cancer-related molecular mechanisms research
- Image Processing Techniques and Applications
- Advanced Image Processing Techniques
- Image Retrieval and Classification Techniques
- Digital Imaging for Blood Diseases
- Topological and Geometric Data Analysis
- Image and Video Quality Assessment
- Video Coding and Compression Technologies
- Image Enhancement Techniques
- Human Pose and Action Recognition
- Machine Learning in Bioinformatics
- Urological Disorders and Treatments
Harvard University
2018-2024
Amazon (United States)
2023
Harvard University Press
2019-2022
Center for Systems Biology
2021-2022
Korea University
2012-2017
An interactive image segmentation algorithm, which accepts user-annotations about a target object and the background, is proposed in this work. We convert into interaction maps by measuring distances of each pixel to annotated locations. Then, we perform forward pass convolutional neural network, outputs an initial map. However, user-annotated locations can be mislabeled result. Therefore, develop backpropagating refinement scheme (BRS), corrects pixels. Experimental results demonstrate that...
A semi-supervised online video object segmentation algorithm, which accepts user annotations about a target at the first frame, is proposed in this work. We propagate labels previous frame to current using optical flow vectors. However, propagation error-prone. Therefore, we develop convolutional trident network (CTN), has three decoding branches: separative, definite foreground, and background decoders. Then, perform Markov random field optimization based on outputs of sequentially carry...
Deep convolutional neural networks (CNNs) have pushed forward the frontier of super-resolution (SR) research. However, current CNN models exhibit a major flaw: they are biased towards learning low-frequency signals. This bias becomes more problematic for image SR task which targets reconstructing all fine details and textures. To tackle this challenge, we propose to improve high-frequency features both locally globally introduce two novel architectural units existing models. Specifically,...
An unsupervised video object segmentation algorithm, which discovers a primary in sequence automatically, is proposed this work. We introduce three energies terms of foreground and background probability distributions: Markov, spatiotemporal, antagonistic energies. Then, we minimize hybrid the to separate from its background. However, energy nonconvex. Therefore, develop alternate convex optimization (ACO) scheme, decomposes nonconvex into two quadratic programs. Moreover, propose...
A graph-based system to simulate the movements and interactions of multiple random walkers (MRW) is proposed in this work. In MRW system, agents traverse a single graph simultaneously. To achieve desired among those agents, restart rule can be designed, which determines distribution each agent according probability distributions all agents. particular, we develop repulsive for data clustering. We illustrate that clustering segment real images reliably. Furthermore, propose novel image...
Facetto is a scalable visual analytics application that used to discover single-cell phenotypes in high-dimensional multi-channel microscopy images of human tumors and tissues. Such represent the cutting edge digital histology promise revolutionize how diseases such as cancer are studied, diagnosed, treated. Highly multiplexed tissue complex, comprising 109 or more pixels, 60-plus channels, millions individual cells. This makes manual analysis challenging error-prone. Existing automated...
Abstract Upcoming technologies enable routine collection of highly multiplexed (20–60 channel), subcellular resolution images mammalian tissues for research and diagnosis. Extracting single cell data from such requires accurate image segmentation, a challenging problem commonly tackled with deep learning. In this paper, we report two findings that substantially improve segmentation using range machine learning architectures. First, unexpectedly find the inclusion intentionally defocused...
A background subtraction algorithm using an encoder-decoder structured convolutional neural network is proposed in this work, order to segment out moving objects from the background. target frame, its previous and a model are concatenated fed into as input. Then, encoder generates highlevel feature vector, decoder converts vector segmentation map, which roughly identifies object regions. Moreover, we develop modeling foreground extraction techniques, exploit contour information. Experimental...
Abstract STUDY QUESTION Can the BlastAssist deep learning pipeline perform comparably to or outperform human experts and embryologists at measuring interpretable, clinically relevant features of embryos in IVF? SUMMARY ANSWER The can measure a comprehensive set interpretable either these features, WHAT IS KNOWN ALREADY Some studies have applied developed ‘black-box’ algorithms predict embryo viability directly from microscope images videos but lack interpretability generalizability. Other...
A primary object discovery (POD) algorithm for a video sequence is proposed in this work, which capable of discovering object, as well identifying noisy frames that do not contain the object. First, we generate proposals each frame. Then, bisect proposal into foreground and background regions, extract features from region. By superposing features, build recurrence model, model. We develop an iterative scheme to refine model evolutionarily using information other models. Finally, evolved...
In this paper, we present the results of MitoEM challenge on mitochondria 3D instance segmentation from electron microscopy images, organized in conjunction with IEEE-ISBI 2021 conference. Our benchmark dataset consists two large-scale volumes, one human and rat cortex tissue, which are 1,986 times larger than previously used datasets. At time paper submission, 257 participants had registered for challenge, 14 teams submitted their results, six participated workshop. Here, eight...
We investigate the impacts of objective functions on performance deep-learning-based prostate magnetic resonance image segmentation. To this end, we first develop a baseline convolutional neural network (BCNN) for segmentation, which consists encoding, bridge, decoding, and classification modules. In BCNN, use 3D layers to consider volumetric information. Also, adopt residual feature forwarding intermediate propagation techniques make BCNN reliably trainable various functions. compare six...
Blastomere instance segmentation is important for analyzing embryos' abnormality. To measure the accurate shapes and sizes of blastomeres, their amodal necessary. Amodal aims to recover an object's complete silhouette even when object not fully visible. For each detected object, previous methods directly regress target mask from input features. However, images under different amounts occlusion should have same output, making it harder train regression model. alleviate problem, we propose...
A novel contour-constrained superpixel (CCS) algorithm is proposed in this work. We initialize superpixels and regions a regular grid then refine the label of each region hierarchically from block to pixel levels. To make boundaries compatible with object contours, we propose notion contour pattern matching formulate an objective function including constraint. Furthermore, extend CCS generate temporal for video processing. labels frame by transferring those previous temporally consistent as...
A real-time video dehazing algorithm, which reduces flickering artifacts and yields high quality output videos, is proposed in this work. Assuming that a scene point highly correlated transmission values between adjacent image frames, we develop the temporal coherence cost. Then, add cost to contrast truncation loss define overall function. By minimizing function, obtain optimal transmission. Moreover, reduce computational complexity facilitate applications, approximate conventional edge...
A tracking-by-segmentation algorithm, which tracks and segments a target object in video sequence, is proposed this paper. In the first frame, we segment out user-annotated bounding box. Then, divide subsequent frames into superpixels. We develop superpixel-wise neural network for tracking-by-segmentation, called TBSNet, extracts multi-level convolutional features of each superpixel yields foreground probability as output. train TBSNet two stages. First, perform offline training to enable...
An efficient coding algorithm for depth map images and videos, based on view synthesis distortion estimation, is proposed in this work. We first analyze how a error related to disparity the vector affects energy spectral density of synthesized color video frequency domain. Based analysis, we propose an estimation technique predict without requiring actual intermediate frames. To encode information efficiently, employ Lagrangian cost function minimize subject constraint transmission bit rate....
ABSTRACT Newly developed technologies have made it feasible to routinely collect highly multiplexed (20-60 channel) images at subcellular resolution from human tissues for research and diagnostic purposes. Extracting single cell data such requires efficient accurate image segmentation, a challenging problem that has recently benefited the use of deep learning. In this paper, we demonstrate two approaches improving tissue segmentation are applicable multiple learning frameworks. The first...
We propose a fast quality metric for depth maps, called (FDQM), which efficiently evaluates the impacts of map errors on qualities synthesized intermediate views in multiview video plus applications. In other words, proposed FDQM assesses view synthesis distortions domain, without performing actual synthesis. First, we estimate at pixel positions, are specified by reference disparities and distorted disparities, respectively. Then, integrate those pixel-wise into an score employing spatial...
A novel quality metric for binary edge maps, called the structural (SEQM), is proposed in this work. First, we define matching cost between an pixel a detected map and its candidate ground-truth map. The includes term, as well positional to measure discrepancy local structures around two pixels. Then, determine optimal pairs of pixels using graph-cut optimization, which smoothness term employed take into account global matching. Finally, sum up costs all index Simulation results demonstrate...