Stereo for Image-Based Rendering using Image Over-Segmentation

In this paper, we propose a stereo method specifically designed for image-based rendering. For effective image-based rendering, the interpolated views need only be visually plausible. The implication is that the extracted depths do not need to be correct, as long as the recovered views appear to be correct. Our stereo algorithm relies on over-segmenting the source images. Computing match values over entire segments rather than single pixels provides robustness to noise and intensity bias. Color-based segmentation also helps to more precisely delineate object boundaries, which is important for reducing boundary artifacts in synthesized views. The depths of the segments for each image are computed using loopy belief propagation within a Markov Random Field framework. Neighboring MRFs are used for occlusion reasoning and ensuring that neighboring depth maps are consistent. We tested our stereo algorithm on several stereo pairs from the Middlebury data set, and show rendering results based on two of these data sets. We also show results for video-based rendering.     

Structure from Motion: Beyond the Epipolar Constraint

The classic approach to structure from motion entails a clear separation between motion estimation and structure estimation and between two-dimensional (2D) and three-dimensional (3D) information. For the recovery of the rigid transformation between different views only 2D image measurements are used. To have available enough information, most existing techniques are based on the intermediate computation of optical flow which, however, poses a problem at the locations of depth discontinuities. If we knew where depth discontinuities were, we could (using a multitude of approaches based on smoothness constraints) accurately estimate flow values for image patches corresponding to smooth scene patches; but to know the discontinuities requires solving the structure from motion problem first. This paper introduces a novel approach to structure from motion which addresses the processes of smoothing, 3D motion and structure estimation in a synergistic manner. It provides an algorithm for estimating the transformation between two views obtained by either a calibrated or uncalibrated camera. The results of the estimation are then utilized to perform a reconstruction of the scene from a short sequence of images.The technique is based on constraints on image derivatives which involve the 3D motion and shape of the scene, leading to a geometric and statistical estimation problem. The interaction between 3D motion and shape allows us to estimate the 3D motion while at the same time segmenting the scene. If we use a wrong 3D motion estimate to compute depth, we obtain a distorted version of the depth function. The distortion, however, is such that the worse the motion estimate, the more likely we are to obtain depth estimates that vary locally more than the correct ones. Since local variability of depth is due either to the existence of a discontinuity or to a wrong 3D motion estimate, being able to differentiate between these two cases provides the correct motion, which yields the least varying estimated depth as well as the image locations of scene discontinuities. We analyze the new constraints, show their relationship to the minimization of the epipolar constraint, and present experimental results using real image sequences that indicate the robustness of the method.     

Automatic View Selection Using Viewpoint Entropy and its Application to Image-Based Modelling

In the last decade a new family of methods, namely Image‐Based Rendering, has appeared. These techniques rely on the use of precomputed images to totally or partially substitute the geometric representation of the scene. This allows to obtain realistic renderings even with modest resources. The main problem is the amount of data needed, mainly due to the high redundancy and the high computational cost of capture. In this paper we present a new method to automatically determine the correct camera placement positions in order to obtain a minimal set of views for Image‐Based Rendering. The input is a 3D polyhedral model including textures and the output is a set of views that sample all visible polygons at an appropriate rate. The viewpoints should cover all visible polygons with an adequate quality, so that we sample the polygons at sufficient rate. This permits to avoid the excessive redundancy of the data existing in several other approaches. We also reduce the cost of the capturing process, as the number of actually computed reference views decreases. The localization of interesting viewpoints is performed with the aid of an information theory‐based measure, dubbed viewpoint entropy. This measure is used to determine the amount of information seen from a viewpoint. Next we develop a greedy algorithm to minimize the number of images needed to represent a scene. In contrast to other approaches, our system uses a special preprocess for textures to avoid artifacts appearing in partially occluded textured polygons. Therefore no visible detail of these images is lost. ACM CSS: I.3.7 Computer Graphics—Three‐Dimensional Graphics and Realism     

Image-Based Rendering Using Parameterized Image Varieties

This paper addresses the problem of characterizing the set of all images of a rigid set of m points and n lines observed by a weak perspective or paraperspective camera. By taking explicitly into account the Euclidean constraints associated with calibrated cameras, we show that the corresponding image space can be represented by a six-dimensional variety embedded in R^2(m+n) and parameterized by the image positions of three reference points. The coefficients defining this parameterized image variety (or PIV for short) can be estimated from a sample of images of a scene via linear and non-linear least squares. The PIV provides an integrated framework for using both point and line features to synthesize new images from a set of pre-recorded pictures (image-based rendering). The proposed technique does not perform any explicit three-dimensional scene reconstruction but it supports hidden-surface elimination, texture mapping and interactive image synthesis at frame rate on ordinary PCs. It has been implemented and extensively tested on real data sets.     

Image-Based Tree Variations

The automatic generation of realistic vegetation closely reproducing the appearance of specific plant species is still a challenging topic in computer graphics. In this paper, we present a new approach to generate new tree models from a small collection of frontal RGBA images of trees. The new models are represented either as single billboards (suitable for still image generation in areas such as architecture rendering) or as billboard clouds (providing parallax effects in interactive applications). Key ingredients of our method include the synthesis of new contours through convex combinations of exemplar countours, the automatic segmentation into crown/trunk classes and the transfer of RGBA colour from the exemplar images to the synthetic target. We also describe a fully automatic approach to convert a single tree image into a billboard cloud by extracting superpixels and distributing them inside a silhouette-defined 3D volume. Our algorithm allows for the automatic generation of an arbitrary number of tree variations from minimal input, and thus provides a fast solution to add vegetation variety in outdoor scenes.     

Computing depth maps from descent images

In the exploration of the planets of our solar system, images taken during a landers descent to the surface of a planet provide a critical link between orbital images and surface images. The descent images not only allow us to locate the landing site in a global coordinate frame, but they also provide progressively higher-resolution maps for mission planning. This paper addresses the generation of depth maps from the descent images. Our approach has two steps, motion refinement and depth recovery. During motion refinement, we use an initial motion estimate in order to avoid the intrinsic motion ambiguity. The objective of the motion-refinement step is to adjust the motion parameters such that the reprojection error is minimized. The depth-recovery step correlates adjacent frames to match pixels for triangulation. Due to the descending motion, the conventional rectification process is replaced by a set of anti-aliasing image warpings corresponding to a set of virtual parallel planes. We demonstrate experimental results on synthetic and real descent images.Received: 7 November 2002, Accepted: 13 September 2004, Published online: 25 February 2005 Correspondence to: Clark F. Olson     

Silhouette‐Aware Warping for Image‐Based Rendering

Image‐based rendering (IBR) techniques allow capture and display of 3D environments using photographs. Modern IBR pipelines reconstruct proxy geometry using multi‐view stereo, reproject the photographs onto the proxy and blend them to create novel views. The success of these methods depends on accurate 3D proxies, which are difficult to obtain for complex objects such as trees and cars. Large number of input images do not improve reconstruction proportionally; surface extraction is challenging even from dense range scans for scenes containing such objects. Our approach does not depend on dense accurate geometric reconstruction; instead we compensate for sparse 3D information by variational image warping. In particular, we formulate silhouette‐aware warps that preserve salient depth discontinuities. This improves the rendering of difficult foreground objects, even when deviating from view interpolation. We use a semi‐automatic step to identify depth discontinuities and extract a sparse set of depth constraints used to guide the warp. Our framework is lightweight and results in good quality IBR for previously challenging environments.     

基于图像的实时漫游

提出了一个点和多边形模型混合的场景表达方式,从而实现了对复杂场景的视点不受限制的实时漫游.它从多幅带深度的参考图像出发,在预处理阶段区别对待场景中平面物体和曲面物体在参考图像中的对应像素.对于在参考图像所占区域较大的空间平面,用传统的多边形模型方式对其进行表达,试图恢复出其平面方程,然后通过采样密度比较和重采样过程,将该空间平面在所有参考图像中的出现合并成一个均匀采样的纹理图像;而对于空间曲面或在参考图像中所占面积较小的平面,使用点的形式对其进行表达,通过采样密度的比较去掉冗余的点,将保留下来的点按其空间位置进行聚类.同时,对于场景中那些不被所有参考图像所拍摄到而在漫游过程中可能形成空洞的部分,提出一个空洞预填补技术,在预处理阶段即对这类空洞进行填补,从而大大减少了漫游阶段出现空洞的几率.在漫游阶段则使用纹理映射和点Warping的方式进行绘制,以充分利用图形硬件的加速功能.     

Lumiproxy: A Hybrid Representation of Image-Based Models

In this paper,we present a hybrid representation of image-based models combining the textured planes and the hierarchical points.Taking a set of depth images as input,our method starts from classifying input pixels into two categories,indicating the planar and non-planar surfaces respectively.For the planar surfaces,the geometric coefficients are reconstructed to form the uniformly sampled textures.For nearly planar surfaces,some textured planes,called lumiproxies, are constructed to represent the equiva...     

Efficient Multi‐image Correspondences for On‐line Light Field Video Processing

Light field videos express the entire visual information of an animated scene, but their shear size typically makes capture, processing and display an off‐line process, i. e., time between initial capture and final display is far from real‐time. In this paper we propose a solution for one of the key bottlenecks in such a processing pipeline, which is a reliable depth reconstruction possibly for many views. This is enabled by a novel correspondence algorithm converting the video streams from a sparse array of off‐the‐shelf cameras into an array of animated depth maps. The algorithm is based on a generalization of the classic multi‐resolution Lucas‐Kanade correspondence algorithm from a pair of images to an entire array. Special inter‐image confidence consolidation allows recovery from unreliable matching in some locations and some views. It can be implemented efficiently in massively parallel hardware, allowing for interactive computations. The resulting depth quality as well as the computation performance compares favorably to other state‐of‐the art light field‐to‐depth approaches, as well as stereo matching techniques. Another outcome of this work is a data set of light field videos that are captured with multiple variants of sparse camera arrays.     

基于图象的室内虚拟漫游系统

基于图象的绘制（Image Based Rendering)作为一种全新的图形绘制方式,以其相对于传统几何绘制而言,具有高效实用的优点,近年来得到了研究人员越来越多的关注,但IBR技术仍存在一些主要难点,如图象的无缝拼接和实时漫游。针对此问题,开发了一个基于部分球面模型的室内虚拟漫游系统。该系统采用自动匹配和人机交互相结合的方法,可以无缝地将多幅照片拼接成一张全景图,同时采用一种改进的基于查找表的算法,实现了固定视点的实时漫游。     

A Network Flow Algorithm for Reconstructing Binary Images from Continuous X-rays

Tomography is a powerful technique to obtain accurate images of the interior of an object in a nondestructive way. Conventional reconstruction algorithms, such as filtered backprojection, require many projections to obtain high quality reconstructions. If the object of interest is known in advance to consist of only a few different materials, corresponding to known image intensities, the use of this prior knowledge in the reconstruction procedure can dramatically reduce the number of required projections. In previous work we proposed a network flow algorithm for reconstructing a binary image defined on a lattice from its projections. In this paper we propose a new algorithm for the reconstruction of binary images that do not have an intrinsic lattice structure and are defined on a continuous domain, from a small number of their projections. Our algorithm relies on the fact that the problem of reconstructing an image from only two projections can be formulated as a network flow problem in a graph. We derive this formulation for parallel beam and fan beam tomography and show how it can be used to compute binary reconstructions from more than two projections. Our algorithm is capable of computing high quality reconstructions from very few projections. We evaluate its performance on both simulated and real experimental projection data and compare it to other reconstruction algorithms.

基于图像的视点变换方程

多年来,造型技术模拟自然界物体形状生成的图像都显得过于规则化和“人工化”。有鉴于此,提出了基于图像的虚拟现实。而完善的虚拟现实系统应该支持多自由度的观察,如视点的移动。视点变换就是求视点移动后原始图像变换生成的新视图。以前给出的变换公式都是在同一坐标系下视点移动后生成的视图。此文给出了在任意坐标系下视点移动后生成新视图的公式,以及具体推导过程。     

Partial Evaluation of Views

Many database applications and environments, such as mediation over heterogeneous database sources and data warehousing for decision support, lead to complex queries. Queries are often nested, defined over previously defined views, and may involve unions. There are good reasons why one might want to remove pieces (sub-queries or sub-views) from such queries: some sub-views of a query may be effectively cached from previous queries, or may be materialized views; some may be known to evaluate empty, by reasoning over the integrity constraints; and some may match protected queries, which for security cannot be evaluated for all users.In this paper, we present a new evaluation strategy with respect to queries defined over views, which we call tuple-tagging, that allows for an efficient removal of sub-views from the query. Other approaches to this are to rewrite the query so the sub-views to be removed are effectively gone, then to evaluate the rewritten query. With the tuple tagging evaluation, no rewrite of the original query is necessary.We describe formally a discounted query (a query with sub-views marked that are to be considered as removed), present the tuple tagging algorithm for evaluating discounted queries, provide an analysis of the algorithm's performance, and present some experimental results. These results strongly support the tuple-tagging algorithm both as an efficient means to effectively remove sub-views from a view query during evaluation, and as a viable optimization strategy for certain applications. The experiments also suggest that rewrite techniques for this may perform worse than the evaluation of the original query, and much worse than the tuple tagging approach.     

Three-Dimensional Reconstruction of Points and Lines with Unknown Correspondence across Images

Three-dimensional reconstruction from a set of images is an important and difficult problem in computer vision. In this paper, we address the problem of determining image feature correspondences while simultaneously reconstructing the corresponding 3D features, given the camera poses of disparate monocular views. First, two new affinity measures are presented that capture the degree to which candidate features from different images consistently represent the projection of the same 3D point or 3D line. An affinity measure for point features in two different views is defined with respect to their distance from a hypothetical projected 3D pseudo-intersection point. Similarly, an affinity measure for 2D image line segments across three views is defined with respect to a 3D pseudo-intersection line. These affinity measures provide a foundation for determining unknown correspondences using weighted bipartite graphs representing candidate point and line matches across different images. As a result of this graph representation, a standard graph-theoretic algorithm can provide an optimal, simultaneous matching and triangulation of points across two views, and lines across three views. Experimental results on synthetic and real data demonstrate the effectiveness of the approach.An erratum to this article can be found at     

Reconstructing 3D human models with a Kinect

Three‐dimensional human model reconstruction has wide applications due to the rapid development of computer vision. The appearance of cheap depth camera, such as Kinect, opens up new horizons for home‐oriented 3D human reconstructions. However, the resolution of Kinect is relatively low, making it difficult to build accurate human models. In this paper, we improve the accuracy of human model reconstruction from two aspects. First, we improve the depth data quality by registering the depth images captured from multi‐views with a single Kinect. The part‐wise registration method and implicit‐surface‐based de‐noising method are proposed. Second, we utilize a statistical human model to iteratively augment and complete the human body information by fitting the statistical human model to the registered depth image. Experimental results and several applications demonstrate the applicability and quality of our system, which can be potentially used in virtual try‐on systems. Copyright © 2015 John Wiley & Sons, Ltd.     

线性多视图重构的新算法

研究了由多幅图像恢复摄像机矩阵和空间物体三维几何形状这一多视图三维重构问题,改进了由Hartley和Rother等人分别给出的基于由无穷远平面诱导的单应进行射影重构的算法,提出了一种新的线性算法,它仅需要空间中3个点在每幅图像上均可见。因为空间中不在同一直线上的3个点恰好确定一个平面,所以它避免了Hartley和Rother等方法中需要确定空间4个点是否共面这一比较棘手的问题。大量实验结果表明,这种方法快速、准确且受噪声影响小。     

Affine shape representation from motion through reference points

This paper shows how an affine representation of spatial configuration is obtained from a pair of projection views. Calibration of cameras and knowledge of the camera's motion are not necessary; however, some preselected reference points and their correspondences are needed. Projective and affine geometry invariants are trickily manipulated to do the affine reconstruction. The method is thus geometrically constructive. When it is compared with the solution proposed in 1989 by J.J. Koenderink and A.J. Van Doorn (Affine Structure from Motion, Technical Report, Utrect University), the method provides a viewpoint-independent affine representation under parallel projections. Further, we investigate the central-projection case in which, with three additional special reference points, the same affine reconstruction can be done. We also discuss some important applications of this viewpoint independence of shape representation.