Search

Ten years into the revival of deep networks and artificial intelligence, we propose a theoretical framework that sheds light on understanding deep networks within a bigger picture of Intelligence in general. We introduce two fundamental principles, Parsimony and Self-consistency, that address two fundamental questions regarding Intelligence: what to learn and how to learn, respectively. We believe the two principles are the cornerstones for the emergence of Intelligence, artificial or natural. While these two principles have rich classical roots, we argue that they can be stated anew in entirely measurable and computable ways. More specifically, the two principles lead to an effective and efficient computational framework, compressive closed-loop transcription, that unifies and explains the evolution of modern deep networks and many artificial intelligence practices. While we mainly use modeling of visual data as an example, we believe the two principles will unify understanding of broad families of autonomous intelligent systems and provide a framework for understanding the brain., Comment: 24 pages, 11 figures. This updated version makes changes in languages and adds a few additional references. This is the final version to be published

Several companies are trying push automatic speech recognition and other technologies past their current limitations.

The lack of fine-grained 3D shape segmentation data is the main obstacle to developing learning-based 3D segmentation techniques. We propose an effective semi-supervised method for learning 3D segmentations from a few labeled 3D shapes and a large amount of unlabeled 3D data. For the unlabeled data, we present a novel multilevel consistency loss to enforce consistency of network predictions between perturbed copies of a 3D shape at multiple levels: point-level, part-level, and hierarchical level. For the labeled data, we develop a simple yet effective part substitution scheme to augment the labeled 3D shapes with more structural variations to enhance training. Our method has been extensively validated on the task of 3D object semantic segmentation on PartNet and ShapeNetPart, and indoor scene semantic segmentation on ScanNet. It exhibits superior performance to existing semi-supervised and unsupervised pre-training 3D approaches. Our code and trained models are publicly available at https://github.com/isunchy/semi_supervised_3d_segmentation., Computational Visual Media 2022. Project page: https://isunchy.github.io/projects/semi_supervised_3d_segmentation.html

Natural language processing (NLP) is a subfield of artificial intelligence that focuses on enabling computers to understand and process human languages. In the last five years, we have witnessed the rapid development of NLP in tasks such as machine translation, question-answering, and machine reading comprehension based on deep learning and an enormous volume of annotated and unannotated data. In this paper, we will review the latest progress in the neural network-based NLP framework (neural NLP) from three perspectives: modeling, learning, and reasoning. In the modeling section, we will describe several fundamental neural network-based modeling paradigms, such as word embedding, sentence embedding, and sequence-to-sequence modeling, which are widely used in modern NLP engines. In the learning section, we will introduce widely used learning methods for NLP models, including supervised, semi-supervised, and unsupervised learning; multitask learning; transfer learning; and active learning. We view reasoning as a new and exciting direction for neural NLP, but it has yet to be well addressed. In the reasoning section, we will review reasoning mechanisms, including the knowledge, existing non-neural inference methods, and new neural inference methods. We emphasize the importance of reasoning in this paper because it is important for building interpretable and knowledge-driven neural NLP models to handle complex tasks. At the end of this paper, we will briefly outline our thoughts on the future directions of neural NLP. Keywords: Natural language processing, Deep learning, Modeling, learning, and reasoning

This article describes the development of Microsoft XiaoIce, the most popular social chatbot in the world. XiaoIce is uniquely designed as an artifical intelligence companion with an emotional connection to satisfy the human need for communication, affection, and social belonging. We take into account both intelligent quotient and emotional quotient in system design, cast human–machine social chat as decision-making over Markov Decision Processes, and optimize XiaoIce for long-term user engagement, measured in expected Conversation-turns Per Session (CPS). We detail the system architecture and key components, including dialogue manager, core chat, skills, and an empathetic computing module. We show how XiaoIce dynamically recognizes human feelings and states, understands user intent, and responds to user needs throughout long conversations. Since the release in 2014, XiaoIce has communicated with over 660 million active users and succeeded in establishing long-term relationships with many of them. Analysis of large-scale online logs shows that XiaoIce has achieved an average CPS of 23, which is significantly higher than that of other chatbots and even human conversations.

Conversational systems have come a long way since their inception in the 1960s. After decades of research and development, we have seen progress from Eliza and Parry in the 1960s and 1970s, to task-completion systems as in the Defense Advanced Research Projects Agency (DARPA) communicator program in the 2000s, to intelligent personal assistants such as Siri, in the 2010s, to today’s social chatbots like XiaoIce. Social chatbots’ appeal lies not only in their ability to respond to users’ diverse requests, but also in being able to establish an emotional connection with users. The latter is done by satisfying users’ need for communication, affection, as well as social belonging. To further the advancement and adoption of social chatbots, their design must focus on user engagement and take both intellectual quotient (IQ) and emotional quotient (EQ) into account. Users should want to engage with a social chatbot; as such, we define the success metric for social chatbots as conversation-turns per session (CPS). Using XiaoIce as an illustrative example, we discuss key technologies in building social chatbots from core chat to visual awareness to skills. We also show how XiaoIce can dynamically recognize emotion and engage the user throughout long conversations with appropriate interpersonal responses. As we become the first generation of humans ever living with artificial intelligenc (AI), we have a responsibility to design social chatbots to be both useful and empathetic, so they will become ubiquitous and help society as a whole.

This paper proposes a movable image-based rendering (M-IBR) system for improving the viewing freedom and environmental modeling capability of conventional static IBR systems. The system supports object-based rendering and 3-D reconstruction capability and consists of three main components. 1) An improved video stabilization method to reduce the shaky motion frequently encountered in movable IBR systems. It employs local polynomial regression (LPR) to automatically select an appropriate bandwidth for smoothing the estimated motion. 2) A novel view synthesis algorithm using a new segmentation and mutual-information (MI)-based algorithm for dense depth map estimation, which relies on segmentation, LPRbased depth map smoothing, and MI-based matching algorithm to iteratively estimate the depth map. The method is very flexible and both semiautomatic and automatic segmentation methods can be employed. They rank fourth and sixth, respectively, in the Middlebury comparison of existing depth estimation methods. This allows high-quality renderings of outdoor scenes with improved mobility/freedom to be obtained. 3) A new 3-D reconstruction algorithm that utilizes the sequential structure-from-motion technique and the dense depth maps estimated previously. It relies on a new iterative point cloud refinement algorithm based on Kalman filter for outlier removal and the segmentation-MI-based algorithm to further refine the correspondences and the projection matrices. The mobility of our system allows us to recover more conveniently 3-D model of static objects from the improved point cloud using a new robust radial basis function-based modeling algorithm to further suppress possible outliers and generate smooth 3-D meshes of objects. Experimental results show that the proposed 3-D reconstruction algorithm significantly reduces the adverse effect of the outliers and produces high-quality renderings using shadow light field and the model reconstructed.

State-of-the-art image retrieval systems achieve scalability by using a bag-of-words representation and textual retrieval methods, but their performance degrades quickly in the face image domain, mainly because they produce visual words with low discriminative power for face images and ignore the special properties of faces. The leading features for face recognition can achieve good retrieval performance, but these features are not suitable for inverted indexing as they are high-dimensional and global and thus not scalable in either computational or storage cost. In this paper, we aim to build a scalable face image retrieval system. For this purpose, we develop a new scalable face representation using both local and global features. In the indexing stage, we exploit special properties of faces to design new component-based local features, which are subsequently quantized into visual words using a novel identity-based quantization scheme. We also use a very small Hamming signature (40 bytes) to encode the discriminative global feature for each face. In the retrieval stage, candidate images are first retrieved from the inverted index of visual words. We then use a new multireference distance to rerank the candidate images using the Hamming signature. On a one millon face database, we show that our local features and global Hamming signatures are complementary--the inverted index based on local features provides candidate images with good recall, while the multireference reranking with global Hamming signature leads to good precision. As a result, our system is not only scalable but also outperforms the linear scan retrieval system using the state-of-the-art face recognition feature in term of the quality.

In this paper, we propose a novel generic image prior-gradient profile prior, which implies the prior knowledge of natural image gradients. In this prior, the image gradients are represented by gradient profiles, which are 1-D profiles of gradient magnitudes perpendicular to image structures. We model the gradient profiles by a parametric gradient profile model. Using this model, the prior knowledge of the gradient profiles are learned from a large collection of natural images, which are called gradient profile prior. Based on this prior, we propose a gradient field transformation to constrain the gradient fields of the high resolution image and the enhanced image when performing single image super-resolution and sharpness enhancement. With this simple but very effective approach, we are able to produce state-of-the-art results. The reconstructed high resolution images or the enhanced images are sharp while have rare ringing or jaggy artifacts.

A new object-based coding system for a class of dynamic image-based representations called plenoptic videos (PVs) is proposed. PVs are simplified dynamic light fields, where the videos are taken at regularly spaced locations along line segments instead of a 2-D plane. In the proposed object-based approach, objects at different depth values are segmented to improve the rendering quality. By encoding PVs at the object level, desirable functionalities such as scalability of contents, error resilience, and interactivity with an individual image-based rendering (IBR) object can be achieved. Besides supporting the coding of texture and binary shape maps for IBR objects with arbitrary shapes, the proposed system also supports the coding of grayscale alpha maps as well as depth maps (geometry information) to respectively facilitate the matting and rendering of the IBR objects. Both temporal and spatial redundancies among the streams in the PV are exploited to improve the coding performance, while avoiding excessive complexity in selective decoding of PVs to support fast rendering speed. Advanced spatial/temporal prediction methods such as global disparity-compensated prediction, as well as direct prediction and its extensions are developed. The bit allocation and rate control scheme employing a new convex optimization-based approach are also introduced. Experimental results show that considerable improvements in coding performance are obtained for both synthetic and real scenes, while supporting the stated object-based functionalities.

We present a real-time algorithm called compensated ray marching for rendering of smoke under dynamic low-frequency environment lighting. Our approach is based on a decomposition of the input smoke animation, represented as a sequence of volumetric density fields, into a set of radial basis functions (RBFs) and a sequence of residual fields. To expedite rendering, the source radiance distribution within the smoke is computed from only the low-frequency RBF approximation of the density fields, since the high-frequency residuals have little impact on global illumination under low-frequency environment lighting. Furthermore, in computing source radiances the contributions from single and multiple scattering are evaluated at only the RBF centers and then approximated at other points in the volume using an RBF-based interpolation. A slice-based integration of these source radiances along each view ray is then performed to render the final image. The high-frequency residual fields, which are a critical component in the local appearance of smoke, are compensated back into the radiance integral during this ray march to generate images of high detail. The runtime algorithm, which includes both light transfer simulation and ray marching, can be easily implemented on the GPU, and thus allows for real-time manipulation of viewpoint and lighting, as well as interactive editing of smoke attributes such as extinction cross section, scattering albedo, and phase function. Only moderate preprocessing time and storage is needed. This approach provides the first method for real-time smoke rendering that includes single and multiple scattering while generating results comparable in quality to offline algorithms like ray tracing.

One of the most important applications in multiview imaging (MVI) is the development of advanced immersive viewing or visualization systems using, for instance, 3DTV. With the introduction of multiview TVs, it is expected that a new age of 3DTV systems will arrive in the near future. Image-based rendering (IBR) refers to a collection of techniques and representations that allow 3-D scenes and objects to be visualized in a realistic way without full 3-D model reconstruction. IBR uses images as the primary substrate. The potential for photorealistic visualization has tremendous appeal, and it has been receiving increasing attention over the years. Applications such as video games, virtual travel, and E-commerce stand to benefit from this technology. This article serves as a tutorial introduction and brief review of this important technology. First the classification, principles, and key research issues of IBR are discussed. Then, an object-based IBR system to illustrate the techniques involved and its potential application in view synthesis and processing are explained. Stereo matching, which is an important technique for depth estimation and view synthesis, is briefly explained and some of the top-ranked methods are highlighted. Finally, the challenging problem of interactive IBR is explained. Possible solutions and some state-of-the-art systems are also reviewed.

In this paper, we study face hallucination, or synthesizing a high-resolution face image from an input low-resolution image, with the help of a large collection of other high-resolution face images. Our theoretical contribution is a two-step statistical modeling approach that integrates both a global parametric model and a local nonparametric model. At the first step, we derive a global linear model to learn the relationship between the high-resolution face images and their smoothed and down-sampled lower resolution ones. At the second step, we model the residue between an original high-resolution image and the reconstructed high-resolution image after applying the learned linear model by a patch-based non-parametric Markov network to capture the high-frequency content. By integrating both global and local models, we can generate photorealistic face images. A practical contribution is a robust warping algorithm to align the low-resolution face images to obtain good hallucination results. The effectiveness of our approach is demonstrated by extensive experiments generating high-quality hallucinated face images from low-resolution input with no manual alignment.

Efficiently simulating large deformations of flexible objects is a challenging problem in computer graphics. In this paper, we present a physically based approach to this problem, using the linear elasticity model and a finite elements method. To handle large deformations in the linear elasticity model, we exploit the domain decomposition method, based on the observation that each sub-domain undergoes a relatively small local deformation, involving a global rigid transformation. In order to efficiently solve the deformation at each simulation time step, we pre-compute the object responses in terms of displacement accelerations to the forces acting on each node, yielding a force-displacement matrix. However, the force-displacement matrix could be too large to handle for densely tessellated objects. To address this problem, we present two methods. The first method exploits spatial coherence to compress the force-displacement matrix using the clustered principal component analysis method; and the second method pre-computes only the force-displacement vectors for the boundary vertices of the sub-domains and resorts to the Cholesky factorization to solve the acceleration for the internal vertices of the sub-domains. Finally, we present some experimental results to show the large deformation effects and fast performance on complex large scale objects under interactive user manipulations.

In this paper, we propose a novel approach for real-time 3-D tracking of object pose from a single camera. We formulate the 3-D pose tracking task in a Bayesian framework which fuses feature correspondence information from both previous frame and some selected key-frames into the posterior distribution of pose. We also developed an inter-frame motion inference algorithm which can get reliable inter-frame feature correspondences and relative pose. Finally, the maximum a posteriori estimation of pose is obtained via stochastic sampling to achieve stable and drift-free tracking. Experiments show significant improvement of our algorithm over existing algorithms especially in the cases of tracking agile motion, severe occlusion, drastic illumination change, and large object scale change

This paper presents a system for capturing and rendering a dynamic image-based representation called the plenoptic video. It is a simplified light field for dynamic environments, where user viewpoints are constrained to the camera plane of a linear array of video cameras. Important issues such as multiple camera calibration, real-time compression, decompression and rendering are addressed. The system consists of a camera array of eight Sony CCX-Z11 CCD cameras and eight Pentium 4 1.8-GHz computers connected together through a 100 Base-T local area network. It is possible to perform software-assisted real-time MPEG-2 compression at a resolution of (720/spl times/480). Using selective transmission, we are able to stream continuously plenoptic video with (256/spl times/256) resolution at a rate of 15 f/s over the network. For rendering from raw data on the hard disk, real-time rendering can be achieved with a resolution of (720/spl times/480) and a rate of 15 f/s. A new compression algorithm using both temporal and spatial predictions is also proposed for the efficient compression of the plenoptic videos. Experimental results demonstrate the usefulness of the proposed parallel processing based system in capturing and rendering high-quality dynamic image-based representations using off-the-shelf equipment, and its potential applications in visualization and immersive television systems.

This paper proposes a novel learning-based approach to synthesizing cursive handwriting of a user's personal handwriting style by combining shape and physical models. In the training process, some sample paragraphs written by a user are collected and these cursive handwriting samples are segmented into individual characters by using a two-level writer-independent segmentation algorithm. Samples for each letter are then aligned and trained using shape models. In the synthesis process, a delta log-normal model based conditional sampling algorithm is proposed to produce smooth and natural cursive handwriting of the user's style from models.

We present a novel technique for large deformations on 3D meshes using the volumetric graph Laplacian. We first construct a graph representing the volume inside the input mesh. The graph need not form a solid meshing of the input mesh's interior; its edges simply connect nearby points in the volume. This graph's Laplacian encodes volumetric details as the difference between each point in the graph and the average of its neighbors. Preserving these volumetric details during deformation imposes a volumetric constraint that prevents unnatural changes in volume. We also include in the graph points a short distance outside the mesh to avoid local self-intersections. Volumetric detail preservation is represented by a quadric energy function. Minimizing it preserves details in a least-squares sense, distributing error uniformly over the whole deformed mesh. It can also be combined with conventional constraints involving surface positions, details or smoothness, and efficiently minimized by solving a sparse linear system.We apply this technique in a 2D curve-based deformation system allowing novice users to create pleasing deformations with little effort. A novel application of this system is to apply nonrigid and exaggerated deformations of 2D cartoon characters to 3D meshes. We demonstrate our system's potential with several examples.

Most modern computers or game consoles are equipped with powerful yet cost-effective graphics processing units (GPUs) to accelerate graphics operations. Though the graphics engines in these GPUs are specially designed for graphics operations, can we harness their computing power for more general nongraphics operations? The answer is positive. In this paper, we present our study on leveraging the GPUs graphics engine to accelerate the video decoding. Specifically, a video decoding framework that involves both the central processing unit (CPU) and the GPU is proposed. By moving the whole motion compensation feedback loop of the decoder to the GPU, the CPU and GPU have been made to work in parallel in a pipelining fashion. Several techniques are also proposed to overcome the GPUs constraints or to optimize the GPU computation. Initial experimental results show that significant speed-up can be achieved by utilizing the GPU power. We have achieved real-time playback of high definition video on a PC with an Intel Pentium III 667-MHz CPU and an nVidia GeForce3 GPU.

This paper proposes a new image-based rendering (IBR) technique called "concentric mosaic" for virtual reality applications. IBR using the plenoptic function is an efficient technique for rendering new views of a scene from a collection of sample images previously captured. It provides much better image quality and lower computational requirement for rendering than conventional three-dimensional (3-D) model-building approaches. The concentric mosaic is a 3-D plenoptic function with viewpoints constrained on a plane. Compared with other more sophisticated four-dimensional plenoptic functions such as the light field and the lumigraph, the file size of a concentric mosaic is much smaller. In contrast to a panorama, the concentric mosaic allows users to move freely in a circular region and observe significant parallax and lighting changes without recovering the geometric and photometric scene models. The rendering of concentric mosaics is very efficient, and involves the reordering and interpolating of previously captured slit images in the concentric mosaic. It typically consists of hundreds of high-resolution images which consume a significant amount of storage and bandwidth for transmission. An MPEG-like compression algorithm is therefore proposed in this paper taking into account the access patterns and redundancy of the mosaic images. The compression algorithms of two equivalent representations of the concentric mosaic, namely the multiperspective panoramas and the normal setup sequence, are investigated. A multiresolution representation of concentric mosaics using a nonlinear filter bank is also proposed.

This paper presents a novel and simple method of analyzing the motion of a large image sequence captured by a calibrated outward-looking video camera moving on a circular trajectory for large-scale environment applications. Previous circular motion algorithms mainly focus on inward-looking turntable-like setups. They are not suitable for outward-looking motion where the conic trajectory of corresponding points degenerates to straight lines. The circular motion of a calibrated camera essentially has only one unknown rotation angle for each frame. The motion recovery for the entire sequence computes only one fundamental matrix of a pair of frames to extract the angular motion of the pair using Laguerre's formula and then propagates the computation of the unknown rotation angles to the other frames by tracking one point over at least three frames. Finally, a maximum-likelihood estimation is developed for the optimization of the whole sequence. Extensive experiments demonstrate the validity of the method and the feasibility of the application in image-based rendering.

Densely-sampled image representations such as the light field or Lumigraph have been effective in enabling photorealistic image synthesis. Unfortunately, lighting interpolation with such representations has not been shown to be possible without the use of accurate 3D geometry and surface reflectance properties. In this paper, we propose an approach to image-based lighting interpolation that is based on estimates of geometry and shading from relatively few images. We decompose light fields captured at different lighting conditions into intrinsic images (reflectance and illumination images), and estimate view-dependent scene geometries using multi-view stereo. We call the resulting representation an Intrinsic Lumigraph. In the same way that the Lumigraph uses geometry to permit more accurate view interpolation, the Intrinsic Lumigraph uses both geometry and intrinsic images to allow high-quality interpolation at different views and lighting conditions. The joint use of geometry and intrinsic images is effective in computing shadow masks for shadow prediction at new lighting conditions. We illustrate our approach with images of real scenes.

Recently, many image-based modeling and rendering techniques have been successfully designed to render photo-realistic images without the need for explicit 3D geometry. However, these techniques (e.g., light field rendering (Levoy, M. and Hanrahan, P., 1996. In SIGGRAPH 1996 Conference Proceedings, Annual Conference Series, Aug. 1996, pp. 31–42) and Lumigraph (Gortler, S.J., Grzeszczuk, R., Szeliski, R., and Cohen, M.F., 1996. In SIGGRAPH 1996 Conference Proceedings, Annual Conference Series, Aug. 1996, pp. 43–54)) may require a substantial number of images. In this paper, we adopt a geometric approach to investigate the minimum sampling problem for light field rendering, with and without geometry information of the scene. Our key observation is that anti-aliased light field rendering is equivalent to eliminating the “double image” artifacts caused by view interpolation. Specifically, we present a closed-form solution of the minimum sampling rate for light field rendering. The minimum sampling rate is determined by the resolution of the camera and the depth variation of the scene. This rate is ensured if the optimal constant depth for rendering is chosen as the harmonic mean of the maximum and minimum depths of the scene. Moreover, we construct the minimum sampling curve in the joint geometry and image space, with the consideration of depth discontinuity. The minimum sampling curve quantitatively indicates how reduced geometry information can be compensated by increasing the number of images, and vice versa. Experimental results demonstrate the effectiveness of our theoretical analysis.

A new approach to computing a panoramic (360 degrees) depth map is presented in this paper. Our approach uses a large collection of images taken by a camera whose motion has been constrained to planar concentric circles. We resample regular perspective images to produce a set of multiperspective panoramas and then compute depth maps directly from these resampled panoramas. Our panoramas sample uniformly in three dimensions: rotation angle, inverse radial distance, and vertical elevation. The use of multiperspective panoramas eliminates the limited overlap present in the original input images and, thus, problems as in conventional multibaseline stereo can be avoided. Our approach differs from stereo matching of single-perspective panoramic images taken from different locations, where the epipolar constraints are sine curves. For our multiperspective panoramas, the epipolar geometry, to the first order approximation, consists of horizontal lines. Therefore, any traditional stereo algorithm can be applied to multiperspective panoramas with little modification. In this paper, we describe two reconstruction algorithms. The first is a cylinder sweep algorithm that uses a small number of resampled multiperspective panoramas to obtain dense 3D reconstruction. The second algorithm, in contrast, uses a large number of multiperspective panoramas and takes advantage of the approximate horizontal epipolar geometry inherent in multiperspective panoramas. It comprises a novel and efficient 1D multibaseline matching technique, followed by tensor voting to extract the depth surface. Experiments show that our algorithms are capable of producing comparable high quality depth maps which can be used for applications such as view interpolation.

We survey the techniques for image-based rendering (IBR) and for compressing image-based representations. Unlike traditional three-dimensional (3-D) computer graphics, in which 3-D geometry of the scene is known, IBR techniques render novel views directly from input images. IBR techniques can be classified into three categories according to how much geometric information is used: rendering without geometry, rendering with implicit geometry (i.e., correspondence), and rendering with explicit geometry (either with approximate or accurate geometry). We discuss the characteristics of these categories and their representative techniques. IBR techniques demonstrate a surprising diverse range in their extent of use of images and geometry in representing 3-D scenes. We explore the issues in trading off the use of images and geometry by revisiting plenoptic-sampling analysis and the notions of view dependency and geometric proxies. Finally, we highlight compression techniques specifically designed for image-based representations. Such compression techniques are important in making IBR techniques practical.

In this paper we present an approach that synthesizes a dynamic sequence from another related sequence, and apply it to a virtual conductor: to synthesize linked figure animation from an input music track. We propose that the mapping between two dynamic sequences can be modeled with a Kernel-based Hidden Markov model, or KHMM. A KHMM is an HMM for which the kernel-based functions are used to model the state observation density of the joint input and output distribution. Specifically, the state observation density is estimated by employing a likelihood-weighted sampling scheme. Our KHMM model is ideal for dynamic sequence synthesis because the global dynamics are learned by the HMM, and subtle details in the dynamic mapping are kept in the kernel-based state density. We demonstrate our virtual conductor by synthesizing extensive animation sequences from input music sequences with different styles and beat patterns.

In this paper, we formulate the stereo matching problem as a Markov network and solve it using Bayesian belief propagation. The stereo Markov network consists of three coupled Markov random fields that model the following: a smooth field for depth/disparity, a line process for depth discontinuity, and a binary process for occlusion. After eliminating the line process and the binary process by introducing two robust functions, we apply the belief propagation algorithm to obtain the maximum a posteriori (MAP) estimation in the Markov network. Other low-level visual cues (e.g., image segmentation) can also be easily incorporated in our stereo model to obtain better stereo results. Experiments demonstrate that our methods are comparable to the state-of-the-art stereo algorithms for many test cases.

In general, mesh smoothing is performed by minimizing the discrete energy function for the surface. One of the major problems in mesh smoothing is to prevent the mesh from shrinking. In this paper, we propose a novel volume constraint to address the shrinking problem in mesh smoothing. Our key observation is that the mesh can be efficiently smoothed patch by patch in a signal processing manner, and then a local volume preserving constraint can be easily imposed to the energy minimization problem associated with the small patch, called the smoothing stencil, so as to effectively avoid the mesh shrinkage. In our implementation, the smoothing stencil is the 1-ring neighboring region of an edge or a triangle. And the constrained minimization problem is solved by a 2-step approximation method for efficiency. A series of examples demonstrate that the proposed smoothing method can be applied to remove noise from a mesh or remove rough detail from an original mesh to generate a smooth model in object reconstruction and geometry modeling.

In this paper, we present a novel image-based rendering technique, which we call manifold hopping. Our technique provides users with perceptually continuous navigation by using only a small number of strategically sampled manifold mosaics or multiperspective panoramas. Manifold hopping has two modes of navigation: moving continuously along any manifold, and discretely between manifolds. An important feature of manifold hopping is that significant data reduction can be achieved without sacrificing output visual fidelity, by carefully adjusting the hopping intervals. A novel view along the manifold is rendered by locally warping a single manifold mosaic using a constant depth assumption, without the need for accurate depth or feature correspondence. The rendering errors caused by manifold hopping can be analyzed in the signed Hough ray space. Experiments with real data demonstrate that we can navigate smoothly in a virtual environment with as little as 88k data compressed from 11 concentric mosaics.

Image-based relighting (IBL) is a technique to change the illumination of an image-based object/scene. In this paper, we define a representation called the reflected irradiance field which records the light reflected from a scene as viewed at a fixed viewpoint as a result of moving a point light source on a plane. It synthesizes a novel image under a different illumination by interpolating and superimposing appropriate recorded samples. Furthermore, we study the minimum sampling problem of the reflected irradiance field, i.e., how many light source positions are needed. We find that there exists a geometry-independent bound for the sampling interval whenever the second-order derivatives of the surface BRDF and the minimum depth of the scene are bounded. This bound ensures that when the novel light source is on the plane, the error in the reconstructed image is controlled by a given tolerance, regardless of the geometry. We also analyze the bound of depth error so that the extra reconstruction error can also be governed when the novel light source is off-plane. Experiments on both synthetic and real surfaces are conducted to verify our analysis.

This paper describes a novel technique (deviation mapping) that generates images with artistic appearances, e.g. spray painted wall, embossment, cashmere painting, etc. Our technique employs a deviation map constructed from a single background image in the image generation process. Instead of recovering the exact geometry from the background image, the deviation map can be regarded as a virtual surface. This virtual surface is then painted with a foreground image and illuminated to generate the final result. Interestingly, the synthesized images exhibit some artistic appearances. Our method is very fast and very simple to implement.

In image based rendering (IBR), a 3D scene is recorded through a set of photos which is then rendered to form novel views. Compression is essential to reduce the huge data amount of IBR. In this paper, we examine three categories of IBR compression algorithms: the block coder, the reference coder and the wavelet coder. We examine both the compression efficiency and the capability to render the compressed IBR bitstream in real-time. It is observed that the block coder consumes the least computation resource, however, its compression ratio is low. The reference coder achieves good compression ratio with reasonable computation complexity. The wavelet coder achieves the best compression ratio.

This paper presents a complete system for constructing panoramic image mosaics from sequences of images. Our mosaic representation associates a transformation matrix with each input image, rather than explicitly projecting all of the images onto a common surface (e.g., a cylinder). In particular, to construct a full view panorama, we introduce a rotational mosaic representation that associates a rotation matrix (and optionally a focal length) with each input image. A patch-based alignment algorithm is developed to quickly align two images given motion models. Techniques for estimating and refining camera focal lengths are also presented. In order to reduce accumulated registration errors, we apply global alignment (block adjustment) to the whole sequence of images, which results in an optimally registered image mosaic. To compensate for small amounts of motion parallax introduced by translations of the camera and other unmodeled distortions, we use a local alignment (deghosting) technique which warps each image based on the results of pairwise local image registrations. By combining both global and local alignment, we significantly improve the quality of our image mosaics, thereby enabling the creation of full view panoramic mosaics with hand-held cameras. We also present an inverse texture mapping algorithm for efficiently extracting environment maps from our panoramic image mosaics. By mapping the mosaic onto an arbitrary texture-mapped polyhedron surrounding the origin, we can explore the virtual environment using standard 3D graphics viewers and hardware without requiring special-purpose players.

This paper presents a feature tracker for long image sequences based on simultaneously estimating die motions and deformations of a collection of adjacent image patches. By sharing common corner nodes, the patches achieve greater stability than independent patch trackers. Modeling full bilinear deformations enables tracking in sequences which have large non-translational motions and/or foreshortening effects. We demonstrate the advantages of our technique with respect to previous algorithms using experimental results.

Presents an approach to free-form object modeling from multiple range images. In most conventional approaches, successive views are registered sequentially. In contrast to the sequential approaches, we propose an integral approach which reconstructs statistically optimal object models by simultaneously aggregating all data from multiple views into a weighted least-squares (WLS) formulation. The integral approach has two components. First, a global resampling algorithm constructs partial representations of the object from individual views, so that correspondence can be established among different views. Second, a weighted least-squares algorithm integrates resampled partial representations of multiple views, using the techniques of principal component analysis with missing data (PCAMD). Experiments show that our approach is robust against noise and mismatch.

Observation-based object modeling often requires integration of shape descriptions from different views. To overcome the problems of errors and their accumulation, we have developed a weighted least-squares (WLS) approach which simultaneously recovers object shape and transformation among different views without recovering interframe motion. We show that object modeling from a range image sequence is a problem of principal component analysis with missing data (PCAMD), which can be generalized as a WLS minimization problem. An efficient algorithm is devised. After we have segmented planar surface regions in each view and tracked them over the image sequence, we construct a normal measurement matrix of surface normals, and a distance measurement matrix of normal distances to the origin for all visible regions over the whole sequence of views, respectively. These two matrices, which have many missing elements due to noise, occlusion, and mismatching, enable us to formulate multiple view merging as a combination of two WLS problems. A two-step algorithm is presented. After surface equations are extracted, spatial connectivity among the surfaces is established to enable the polyhedral object model to be constructed. Experiments using synthetic data and real range images show that our approach is robust against noise and mismatching and generates accurate polyhedral object models. >

In space application, robot system are subject to unknown or unmodeled dynamics, for example, in the tasks of transporting an unknown payload or catching an unmodeled moving object. We discuss the parameterization problem in dynamic structure and adaptive control of a space robot system with an attitude-controlled base to which the robot is attached. We first derive the system kinematic and dynamic equations based on Lagrangian dynamics and the linear momentum conservation law. Based on the dynamic model developed, we discuss the problem of linear parameterization in term of dynamic parameters, and find that in joint space, the dynamics can be linearized by a set of combined dynamic parameters; however, in inertial space linear parameterization is impossible in general. Then we propose an adaptive control scheme in joint space, and present a simulation study to demonstrate its effectiveness and computational procedure. Because most takes are specified in inertial space instead of joint space, we discuss the issues associated to adaptive control in inertial space and identify two potential problem: unavailability of joint trajectory because the mapping from inertial space trajectory is dynamic-dependent and subject to uncertainty; and nonlinear parameterization in inertial space. We approach the problem by making use of the proposed joint space adaptive controller and updating the joint trajectory by the estimated dynamic parameters and given trajectory in inertial space. >

In this paper, we study the salient object detection problem for images. We formulate this problem as a binary labeling task where we separate the salient object from the background. We propose a set of novel features, including multiscale contrast, center-surround histogram, and color spatial distribution, to describe a salient object locally, regionally, and globally. A conditional random field is learned to effectively combine these features for salient object detection. Further, we extend the proposed approach to detect a salient object from sequential images by introducing the dynamic salient features. We collected a large image database containing tens of thousands of carefully labeled images by multiple users and a video segment database, and conducted a set of experiments over them to demonstrate the effectiveness of the proposed approach.

This paper proposes an efficient algorithm for the compression, prioritized transmission and progressive rendering of circular light field (CLF) for ancient Chinese artifacts. It employs wavelet coder to achieve spatial scalability and divide the compressed data into a lower resolution base layer and an additional enhancement layer. The enhancement layer is coded as in JPEG2000 into packets where the base-layer is coded using disparity compensation prediction (DCP). The frame structure is designed to provide efficient access to the compressed data in order to support selective transmission and decoding. The depth and alpha maps are coded analogously. A prioritized transmission scheme which support interactive progressive rendering is also proposed to further reduce the latency and response time of rendering.

Image-based rendering (IBR) is an emerging and promising technology for photo-realistic rendering of scenes and objects from a collection of densely sampled images and videos. This paper proposes an image-based approach to the rendering and multi-view display of ancient Chinese artifacts for cultural heritage preservation. A multiple-camera circular array was constructed to record images of the artifacts. Novel techniques for segmenting and rendering new views of the artifacts from the sampled images are developed. The multiple views so synthesized enable the ancient artifacts to be displayed in modern multi-view displays and conventional stereo systems. Several collections from the University Museum and Art Gallery at the University of Hong Kong are captured and excellent rendering results are obtained.

The quality and speed of most texture synthesis algorithms depend on a 2D input sample that is small and contains enough texture variations. However, little research exists on how to acquire such sample. For homogeneous patterns this can be achieved via manual cropping, but no adequate solution exists for inhomogeneous or globally varying textures, i.e. patterns that are local but not stationary, such as rusting over an iron statue with appearance conditioned on varying moisture levels. We present inverse texture synthesis to address this issue. Our inverse synthesis runs in the opposite direction with respect to traditional forward synthesis: given a large globally varying texture, our algorithm automatically produces a small texture compaction that best summarizes the original. This small compaction can be used to reconstruct the original texture or to re-synthesize new textures under user-supplied controls. More important, our technique allows real-time synthesis of globally varying textures on a GPU, where the texture memory is usually too small for large textures. We propose an optimization framework for inverse texture synthesis, ensuring that each input region is properly encoded in the output compaction. Our optimization process also automatically computes orientation fields for anisotropic textures containing both low- and high-frequency regions, a situation difficult to handle via existing techniques.

We present a simple interactive approach to specify 3D shape in a single view using "shape palettes". The interaction is as follows: draw a simple 2D primitive in the 2D view and then specify its 3D orientation by drawing a corresponding primitive on a shape palette . The shape palette is presented as an image of some familiar shape whose local 3D orientation is readily understood and can be easily marked over. The 3D orientation from the shape palette is transferred to the 2D primitive based on the markup. As we will demonstrate, only sparse markup is needed to generate expressive and detailed 3D surfaces. This markup approach can be used to model freehand 3D surfaces drawn in a single view, or combined with image-snapping tools to quickly extract surfaces from images and photographs.

We study visual attention by detecting a salient object in an input image. We formulate salient object detection as an image segmentation problem, where we separate the salient object from the image background. We propose a set of novel features including multi-scale contrast, center-surround histogram, and color spatial distribution to describe a salient object locally, regionally, and globally. A conditional random field is learned to effectively combine these features for salient object detection. We also constructed a large image database containing tens of thousands of carefully labeled images by multiple users. To our knowledge, it is the first large image database for quantitative evaluation of visual attention algorithms. We validate our approach on this image database, which is public available with this paper.