CVGIP: Image Understanding最新文献

An approach to labeling the components of human faces from range images is proposed. The components of interest are those humans usually find significant for recognition. To cope with the nonrigidity of faces, an entirely qualitative approach is used. The preprocessing stage employs a multistage diffusion process to identify convexity and concavity points. These points are grouped into components and qualitative reasoning about possible interpretations of the components is performed. Consistency of hypothesized interpretations is carried out using context-based reasoning. Experimental results on real range images of several faces are provided.

In this paper, we propose a scheme for 3D model construction by fusing heterogeneous sensor data. The proposed scheme is intended for use in an environment where multiple, heterogeneous sensors operate asynchronously. Surface depth, orientation, and curvature measurements obtained from multiple sensors and vantage points are incorporated to construct a computer description of the imaged object. The proposed scheme uses Kalman filter as the sensor data integration tool and hierarchical spline surface as the recording data structure. Kalman filter is used to obtain statistically optimal estimates of the imaged surface structure based on possibly noisy sensor measurements. Hierarchical spline surface is used as the representation scheme because it maintains high-order surface derivative continuity, may be adaptively refined, and is storage efficient. We show in this paper how these mathematical tools can be used in designing a modeling scheme to fuse heterogeneous sensor data.

A new technique for computing optical flow from an extended sequence (containing more than two images) of image frames is proposed. The proposed technique explicitly utilizes the additional information present in the extended frame sequence by utilizing the smoothness of trajectory of intensity points as a constraint. Importance of trajectory smoothness of intensity points is established and its mathematical formulation is derived in terms of three components. Discontinuities in the trajectories are also modeled by a field of binary elements. Estimation of the unknown optical flow field together with the discontinuities is formulated as a Bayesian maximum a posteriori (MAP) probability estimation problem. The conditional probability of the unknown velocity and discontinuity fields, given the observed image sequence, is computed based on the trajectory and spatial smoothness model. The correspondingdistribution is shown to be a Gibbs distribution (equivalently a Markov random field). The "most probable velocity state" is then found by a stochastic relaxation algorithm. Experimental results with both synthetic and real image sequences are presented to demonstrate the efficacy of the method. In cases where ground truth is known, error estimates for the proposed technique are provided and compared with that for other well-known methods.

We estimate optical flow from a sequence of 2-D images by computing the velocity field along moving contours in the scene. This new approach is different from others in that it combines displacements computed by feature matching with a smoothness constraint on the second derivative of velocity. First, we use our previously reported relaxation matching technique to find correspondences between contour features in adjacent image frames. Displacements for discrete points along the contours are interpolated from the magnitudes and directions of neighboring matched points. The displacements so-computed are used as initial estimates for the velocity (magnitude and direction) along contours. The final estimated velocities are required to yield components which are close in a least-squares sense to these initial velocity magnitudes, when projected along the same directions. We also constrain the second derivative of velocity to be a minimum when integratedalong the contour, leading to a unique solution for the motion of a straight line undergoing an affine transformation. The second-derivative constraint gives better results than the first-derivative constraint in this case. Our method also gives better results for most second-order flows. In cases where it does not, a combination of first- and second-derivative constraints can be used. Computation of velocities at discrete points along the contour is achieved by solving linear equations via the conjugate gradient algorithm. The image flow technique is applied to examples of rigid and nonrigid motion.

Address block location on mail pieces is an important task in postal automation. Unlike personal or business letters which have high degree of global spatial structure among a limited number of entities, mail pieces of magazines usually have an address block printed on a white label which can be pasted in an arbitrary position. Graphics and other printed text on magazine covers also make the address block location problem complicated. In this paper, we present a simple method for automatically locating the address block on color magazine covers based on both color and texture analysis. First, we use a simple color thresholding technique to extract white regions which may contain an address block. Then, a texture segmentation method based on Gabor filters is used to find text regions (including the address) inside the white regions. These text regions are candidates for the address block. Simple heuristics are used to identify the correct address block among these candidates. This method is invariant to rotation and scale of the magazine cover. Experimental results on low resolution (50 dpi) images of several magazine covers are provided to demonstrate the applicability of our method.

This paper describes a physically inspired method for the recovery of the surface of 3D solid objects from sparse data. The method is based on a model of closed elastic thin surface under the action of radial springs which can be considered as the analogous, in spherical coordinates, to the well-known thin plate model. The model is a representation for whole-body surfaces which has the degrees of freedom for representing fine details. We formulate the surface recovery problem as the problem of minimizing a non-quadratic energy functional. In the hypothesis of small deformations, this functional is approximated with a quadratic one which is then discretized with the finite element method. We provide steepest-descent-like algorithms both for the case of small deformations and for that of large ones. Then we introduce a representation of our model in terms of its free deformation modes. This representation is extremely concise and is therefore suited for shape analysis and recognition tasks. Finally, we report on the results of experiments with synthetic and real data which show the performance of the method