We present an end-to-end system that goes from video sequences to high resolution, editable, dynamically controllable face models. The capture employs synchronized cameras and structured light projectors record videos of a moving multiple viewpoints. A novel spacetime stereo algorithm is introduced compute depth maps accurately overcome over-fitting deficiencies in prior work. new template fitting tracking procedure fills missing data yields point correspondence across the entire sequence...

This paper presents a color structured light technique for recovering object shape from one or more images. The works by projecting pattern of stripes alternating colors and matching the projected transitions with observed edges in image. correspondence problem is solved using novel, multi-pass dynamic programming algorithm that eliminates global smoothness assumptions strict ordering constraints present previous formulations. resulting approach suitable generating both high-speed scans...

We present a novel algorithm for view synthesis that utilizes soft 3D reconstruction to improve quality, continuity and robustness. Our main contribution is the formulation of representation preserves depth uncertainty through each stage rendering. show this beneficial throughout pipeline. During synthesis, it provides model scene geometry across synthesized views robustness uncertainty. reconstruction, same robust estimates visibility can be applied iteratively estimation around object...

This paper extends the traditional binocular stereo problem into spacetime domain, in which a pair of video streams is matched simultaneously instead matching pairs images frame by frame. Almost any existing algorithm may be extended this manner simply replacing image term with term. By utilizing both spatial and temporal appearance variation, modification reduces ambiguity increases accuracy. Three major applications for are proposed paper. First, serves as general framework structured...

This paper presents a novel approach for reconstructing free-form, texture-mapped, 3D scene models from single painting or photograph.Given sparse set of user-specified constraints on the local shape scene, smooth surface that satisfies is generated.This problem formulated as constrained variational optimization problem.In contrast to previous work in view reconstruction, our technique enables high quality reconstructions free-form curved surfaces with arbitrary reflectance properties.A key...

Cost aggregation is one of the key steps in stereo matching problem. In order to improve accuracy, we propose a cost-aggregation method that can embed minimum spanning tree (MST)-based support region filtering into PatchMatch 3D label search rather than aggregating on fixed size patches. However, directly combining and MST not straightforward, due extremely high complexity. Thus, develop multiple structures for cost plenty labels, design tree-level random strategy find possible labels each...

China's soybean supply and demand are seriously imbalanced. It is crucial to improve the level of breeding. Hundred-grain weight one most essential phenotypic parameters for crop Accurate seed counting a key step 100-grain weight. There several methods, which have their own limitations way or other. Among these, manual time-consuming, electronic automatic counter devices expensive speed very slow, traditional digital image processing techniques not suitable based on individual pod images....

Effective and efficient fruit detection is considered crucial for designing automated robot (AuRo) yield estimation, disease control, harvesting, sorting, grading. Several schemes AuRo have been developed during the last decades. However, conventional methods are deficient in real-time response, accuracy, extensibility. This paper proposes an improved multi-task cascaded convolutional network-based intelligent method. method has capability to make work real time with high accuracy. Moreover,...

We present an end-to-end system that goes from video sequences to high resolution, editable, dynamically controllable face models. The capture employs synchronized cameras and structured light projectors record videos of a moving multiple viewpoints. A novel spacetime stereo algorithm is introduced compute depth maps accurately overcome over-fitting deficiencies in prior work. new template fitting tracking procedure fills missing data yields point correspondence across the entire sequence...

Abstract Compared with integer order chaotic systems, fractional systems can reflect natural phenomena more accurately, which are suitable for cryptosystems. In to explore the application of system in cryptography, a novel hyperchaotic is constructed and implemented on DSP platform. More progressively, based Adomian decomposition method, dynamic behavior studied by phase diagram, bifurcation Lyapunov exponent spectrum spectral entropy (SE) complexity. It found that each parameter have large...

Abstract This paper presents a novel approach for reconstructing free‐form, texture‐mapped, 3D scene models from single painting or photograph. Given sparse set of user‐specified constraints on the local shape scene, smooth surface that satisfies is generated. problem formulated as constrained variational optimization problem. In contrast to previous work in single‐view reconstruction, our technique enables high‐quality reconstructions free‐form curved surfaces with arbitrary reflectance...

Fuji apples are one of the most important and popular economic crops worldwide in fruit industry. Nowadays, there is a huge imbalance between urgent demand precise automated sorting models ripeness grades due to increasing consumption levels limitations existing methods. In this regard, paper presents novel CNN-based fine-grained lightweight architecture for task apple maturity classification (FGAL-MC). Our proposed FGAL-MC has three advantages compared with related previous research works....

This paper presents a novel approach for estimating parameters MRF-based stereo algorithms. is based on new formulation of as maximum posterior (MAP) problem, in which both disparity map and MRF are estimated from the pair itself. We present an iterative algorithm MAP estimation that alternates between while fixing parameters. The include robust truncation thresholds, data neighborhood terms, well regularization weight. weight can be either constant whole image, or spatially-varying,...

This paper presents an automatic method for creating a collage from collection of photos scene taken different viewpoints. The is constructed by aligning the images (in terms their positions, rotations and scales) using least-squares formulation. We have developed graph-based optimization algorithm layering so as to minimize fragmentation collage. A can be displayed with opaque layers, transparent or blended image boundaries. viewed piece-wise perspective representation visible seams. has...

In this paper, we present a new framework for non-rigid structure from motion (NRSFM) that simultaneously addresses three significant challenges: severe occlusion, perspective camera projection, and large non-linear deformation. We introduce concept called model graph, which greatly reduces the computational cost of discovering groups input images depict consistent 3D shapes. A is constructed each image by traversing graph along multiple evolutionary paths. compressive shape representation...

Environment depth estimation by fusing camera and radar enables a broad spectrum of applications such as autonomous driving, environmental perception, context-aware localization navigation. Various pioneering approaches have been proposed to achieve accurate dense integrating vision LiDAR through deep learning. However, due the challenges sparse sampling in-vehicle LiDARs, high ground-truth annotation overhead, severe dynamics in real environments, existing solutions not yet achieved...

Generated topic label, an alternative representation of topics learned by model, is widely used to help the user interpret more efficiently. A major challenge now label a discovered accurately in objective way. This article introduces novel graph-based ranking model (TLRank), find meaningful with high Relevance, Coverage, and Discrimination. The applies specific strategy that suppresses or enhances matrix transition probability according textual similarity between vertices (sentences)...

In order to improve the accuracy of image segmentation, an improved adaptive level set method is proposed based on evolution without re-initialization and distance preserving method. A new definition weight coefficient in equations main innovation this paper. The can detect certain object boundaries, interior exterior contours object, edges multi-objects weak boundaries by synthetic real images numerical experiments. Numerical results show that has faster segmentation speed higher compared...

Pedestrian detection is a key problem for automatic driving, and the results have been improved significantly via deep convolutional networks. However, there still room to improve performance of pedestrian by carefully dealing with some critical issues. To take advantages more discriminative information detection, we propose novel architecture auto-choose semantic as well specific among feature maps at different levels integrate valuable in multi-scales. Particularly, our consists...

Depth estimation is one of the bottleneck parts for harvesting robots to determine whether operation grasping or picking succeeds not directly. This paper proposed a novel disparity completion method combined with bilateral filtering and pyramid fusion improve issues incorrect outputs due missed wrong matching when achieving 3D position from 2D images in open-world environments. Briefly, our has two significant advantages general. Firstly, occlusion between leaves, branches, fruits universal...

To meet the growing demand for point cloud compression, MPEG is developing a compression standard called as G-PCC. In G-PCC, upsampled transform domain prediction (UTDP) used to improve attribute coding performance. However, only attributes in previous level can be predict of sub-blocks UTDP, which limits efficiency UTDP. address this limitation, we propose method peer-UTDP UTDP by using peer neighbors paper. With peer-UTDP, co-plane or co-line same that sub-block upsampling process....

Large oblique stereo images are of great interest because they have a large coverage and high reconstruction precision. However, severe distortions more likely to occur in this type image. This can make conventional algorithms inaccurate, computationally expensive, invalid. We present new robust quasidense image matching algorithm for images. Our be divided into two steps. First, we find sufficient number highly accurate seed matches that uniformly distributed by integrating complementary...