A novel method is proposed for the problem of frame-to-frame correspondence search in video sequences. The method, based on hashing of low-dimensional image descriptors, establishes dense correspondences and allows la...
详细信息
A novel method is proposed for the problem of frame-to-frame correspondence search in video sequences. The method, based on hashing of low-dimensional image descriptors, establishes dense correspondences and allows large motions. All image pixels are considered for matching, the notion of interest points is reviewed. In our formulation, points of interest are those that can be reliably matched. Their saliency depends on properties of the chosen matching function and on actual image content. Both computational time and memory requirements of the correspondence search are asymptotically linear in the number of image pixels, irrespective of correspondence density and of image content. All steps of the method are simple and allow for a hardware implementation. Functionality is demonstrated on sequences taken from a vehicle moving in an urban environment.
This paper discusses fast and accurate methods to solve total variation (TV) models on the graphics processing unit (GPU). We review two prominent models incorporating TV regularization and present different algorithm...
详细信息
This paper discusses fast and accurate methods to solve total variation (TV) models on the graphics processing unit (GPU). We review two prominent models incorporating TV regularization and present different algorithms to solve these models. We mainly concentrate on variational techniques, i.e. algorithms which aim at solving the Euler Lagrange equations associated with the variational model. We then show that particularly these algorithms can be effectively accelerated by implementing them on parallel architectures such as GPUs. For comparison we chose a state-of-the-art method based on discrete optimization techniques. We then present the results of a rigorous performance evaluation including 2D and 3D problems. As a main result we show that the our GPU based algorithms clearly outperform discrete optimization techniques in both speed and maximum problem size.
This paper presents a new method for reconstructing rectilinear buildings from single images under the assumption of flat terrain. An intuition of the method is that, given an image composed of rectilinear buildings, ...
详细信息
This paper presents a new method for reconstructing rectilinear buildings from single images under the assumption of flat terrain. An intuition of the method is that, given an image composed of rectilinear buildings, the 3D buildings can be geometrically reconstructed by using the image only. The recovery algorithm is formulated in terms of two objective functions which are based on the equivalence between the vector normal to the interpretation plane in the image space and the vector normal to the rotated interpretation plane in the object space. These objective functions are minimized with respect to the camera pose, the building dimensions, locations and orientations to obtain estimates for the structure of the scene. The method potentially provides a solution for large-scale urban modelling using aerial images, and can be easily extended to deal with piecewise planar objects in a more general situation.
Many perceptual grouping algorithms depend on parameters one way or another. It is always difficult to set these parameters appropriately for a wide range of input images, and parameters tend to be tuned to a small se...
详细信息
Many perceptual grouping algorithms depend on parameters one way or another. It is always difficult to set these parameters appropriately for a wide range of input images, and parameters tend to be tuned to a small set of test cases. Especially certain thresholds often seem unavoidable to limit search spaces in order to obtain reasonable runtime complexity. Furthermore early pruning of less salient hypotheses is often necessary to avoid exponential growth of the number of hypotheses. In the presented work we show how the adoption of a simple anytime algorithm, i. e. an algorithm which returns the best answer possible when interrupted and may improve on the answer if allowed to run longer, for finding closed convex polygons eliminates the need for parameter tuning. Furthermore it quite naturally allows the incorporation of attentional mechanisms into the grouping process.
We propose a Bayesian network (BN) model to integrate multiple contextual information and the image measurements for image segmentation. The BN model systematically encodes the contextual relationships between regions...
详细信息
We propose a Bayesian network (BN) model to integrate multiple contextual information and the image measurements for image segmentation. The BN model systematically encodes the contextual relationships between regions, edges and vertices, as well as their image measurements with uncertainties. It allows a principled probabilistic inference to be performed so that image segmentation can be achieved through a most probable explanation (MPE) inference in the BN model. We have achieved encouraging results on the horse images from the Weizmann dataset. We have also demonstrated the possible ways to extend the BN model so as to incorporate other contextual information such as the global object shape and human intervention for improving image segmentation. Human intervention is encoded as new evidence in the BN model. Its impact is propagated through belief propagation to update the states of the whole model. From the updated BN model, new image segmentation is produced.
This work targets the design of customized accelerators for image registration algorithms, which are required for many important computervision applications. By capturing key, domain-specific characteristics of appli...
详细信息
This work targets the design of customized accelerators for image registration algorithms, which are required for many important computervision applications. By capturing key, domain-specific characteristics of application structure, signal-processing-oriented models of computation provide a valuable foundation for structured development of efficient image registration accelerators. Building upon the meta-modeling framework of homogeneous parameterized dataflow, we develop in this paper an approach for automatically generating streamlined implementations of image registration algorithms according to performance metrics such as image size, area and overall processing speed. Results from hardware synthesis demonstrate the efficiency of our methods. Our approach provides designers an effective way to explore different architectures, and systematically provide acceleration for high-performance nonrigid image registration based on a variety of requirements. Our dataflow-based framework can be adapted to explore different architectures for other kinds of image processing algorithms as well.
Contour features are re-emerging in the categorization community as it moves from appearance back to shape. However, the classical assumption of one-to-one correspondence between an extracted image contour and a model...
详细信息
Contour features are re-emerging in the categorization community as it moves from appearance back to shape. However, the classical assumption of one-to-one correspondence between an extracted image contour and a model contour constrains category models to be highly brittle, offering little abstraction between image and model. Moreover, todaypsilas contour-based models are category-specific, offering no mechanism for contour grouping and abstraction in the absence of an object prior. We present a novel framework for recovering a set of abstract parts from a multi-scale contour image. Given a user-specified part vocabulary and an image to be analyzed, the system covers the image with abstract part models drawn from the vocabulary. More importantly, correspondence between image contours and part contours is many-to-one, yielding a powerful shape abstraction mechanism. We illustrate the strengths and weaknesses of this work in progress on a set of anecdotal scenes.
Non-rigid surface registration, particularly registration of human faces, finds a wide variety of applications in computervision and graphics. We present a new automatic surface registration method which utilizes bot...
详细信息
Non-rigid surface registration, particularly registration of human faces, finds a wide variety of applications in computervision and graphics. We present a new automatic surface registration method which utilizes both attraction forces originating from geometrical and textural similarities, and stresses due to non-linear elasticity of the surfaces. Reference and target surfaces are first mapped onto their feature image planes, then these images are registered by subjecting them to local deformations, and finally 3D correspondences are established. Surfaces are assumed to be elastic sheets and are represented by triangular meshes. The internal elastic forces act as a regularizer in this ill-posed problem. Furthermore, the non-linear elasticity model allows us to handle large deformations, which can be essential, for instance, for facial expressions. The method has been tested successfully on 3D scanned human faces, with and without expressions. The algorithm runs quite efficiently using a multiresolution approach.
3D imaging systems provide valuable information for autonomous robot navigation based on landmark detection in pipelines. This paper presents a method for using a time-of-flight (TOF) camera for detection and tracking...
详细信息
3D imaging systems provide valuable information for autonomous robot navigation based on landmark detection in pipelines. This paper presents a method for using a time-of-flight (TOF) camera for detection and tracking of pipeline features such as junctions, bends and obstacles. Feature extraction is done by fitting a cylinder to images of the pipeline. Data in captured images appear to take a conic rather than cylindrical shape, and we adjust the geometric primitive accordingly. Pixels deviating from the estimated cylinder/cone fit are grouped into blobs. Blobs fulfilling constraints on shape and stability over time are then tracked. The usefulness of TOF imagery as a source for landmark detection and tracking in pipelines is evaluated by comparison to auxiliary measurements. Experiments using a model pipeline and a prototype robot show encouraging results.
Regularization is an important aspect in high angular resolution diffusion imaging (HARDI), since, unlike with classical diffusion tensor imaging (DTI), there is no a priori regularity of raw data in the co-domain, i....
详细信息
Regularization is an important aspect in high angular resolution diffusion imaging (HARDI), since, unlike with classical diffusion tensor imaging (DTI), there is no a priori regularity of raw data in the co-domain, i.e. considered as a multispectral signal for fixed spatial position. HARDI preprocessing is therefore a crucial step prior to any subsequent analysis, and some insight in regularization paradigms and their interrelations is compulsory. In this paper we posit a codomain scale space regularization paradigm that has hitherto not been applied in the context of HARDI. Unlike previous (first and second order) schemes it is based on infinite order regularization, yet can be fully operationalized. We furthermore establish a closed-form relation with first order Tikhonov regularization via the Laplace transform.
暂无评论