Graphical Models and Image Processing最新文献

Finitary 1-simply connected digital spaces are discrete analogs of the important simply connected spaces in classical topology (i.e., connected spaces in which every loop can be continuously pulled to a point without leaving the space). Loosely speaking, 1-simply connected digital spaces are graphs in which there are no holes larger than a triangle. Many spaces previously studied in digital topology and geometry are instances of this concept. Boundaries in pictures defined over finitary 1-simply connected digital spaces have some desirable general properties; for example, they partition the space into a connected interior and a connected exterior. There is a “one-size-fits-all” algorithm which, given a picture over a finitary 1-simply connected digital space and a boundary face, will return the set of all faces in that boundary, provided only that this set is finite; the proof of correctness of this algorithm is an immediate consequence of the general properties of such spaces.

This paper presents pixel-based and region-based parametric motion segmentation methods for robust motion segmentation with the goal of aligning motion boundaries with those of real objects in a scene. We first describe a two-step iterative procedure for parametric motion segmentation by either motion-vector or motion-compensated intensity matching. We next present a region-based extension of this method, whereby all pixels within a predefined spatial region are assigned the same motion label. These predefined regions may be fixed- or variable-size blocks or arbitrary-shaped areas defined by color or texture uniformity. A particular combination of these pixel-based and region-based methods is then proposed as a complete algorithm to obtain the best possible segmentation results on a variety of image sequences. Experimental results showing the benefits of the proposed scheme are provided.

Recognition of documents of poor image quality is a challenging and important problem from a practical point of view. In traditional approaches, features such as center lines of strokes or contours are extracted from binary images obtained by thresholding the gray-scale intensity images. Wang and Pavlidis (IEEE Trans. Pattern Anal. Machine Intell. 15(10), 1993, 1053–1067) have recently pointed out that effective features for recognition should be extracted directly from original gray-scale intensity images in order to avoid a significant amount of information loss caused by binarization. In this paper, a novel method is presented for extracting closed boundaries of document components such as characters and symbols directly from gray-scale document images, based on the surface data structures and structural features. The gray-scale document image can be treated as a surface defined over a two-dimensional space by regarding intensity values associated with pixels as height. This method is based on a simple model that assumes a closed boundary of document components can be approximated as a series of horizontal (parallel to the image plane) line segments and can be extracted by linking surface components with steep gradients based on configurations of intersections of horizontal planes and surface components. Furthermore, the gray-scale image can be converted into a binary image based on extracted boundaries so that any recognition system can accept output of the proposed algorithm as input. The performance of the proposed algorithm is compared with some binarization algorithms based on global and local thresholding of intensity values and is shown to be effective for improving recognition accuracy for very poor quality data.

Utilizing information theory and considering image segmentation from a communication perspective, the image segmentation process is interpreted as a data processing step that operates on a gray-scale image and produces a segmented image. It is shown that the segmented image contains a certain amount of information about the scene, which is defined assegmented image information(SII). It is proposed that the SII should be maximized when an image is thresholded, and this is known as themaximum segmented image information(MSII) thresholding criterion. The MSII thresholding criterion possesses better properties as compared with theminimum error(MINE) and theuniform error(UNFE) thresholding criteria. Based on the MSII thresholding criterion, an MSII thresholding algorithm is proposed for the thresholding of real images. The MSII thresholding algorithm is evaluated against several well-known thresholding algorithms. The good thresholding results of both synthetic and real images confirm the capabilities of the proposed MSII thresholding algorithm.

Visibility problems are investigated using reconfigurable meshes. A number of algorithms are proposed on the architecture for visibility computation in two and three dimensions. We show that visibility of a total ofndisjoint edges in the plane can be computed inO(1) time on ann×nmesh. The result is optimal in the word model of VLSI. For the case that the edges are not disjoint, the problem is shown to be solvable inO(1) time by using a mesh of slightly larger size or in slightly more time on ann×nmesh. We also present hidden-line and surface elimination algorithms that run on ann×n×nmesh for a set of disjoint triangles in 3-space containing a total ofnvertices inO(1) time andO(k) time, respectively, where 0 ≤k<nis an output-dependent parameter.

Recognizing the construction methods of (piecewise) polynomial or rational curves and surfaces is of great importance, e.g., for geometrical data exchange between two different modeling systems. We formulateintrinsicconditions that are parameterization independent whenever possible. These conditions can detect: (i) whether a curve segment is a line, a circle, or a planar curve; (ii) whether a surface patch is a plane, a sphere, a cylinder, or a cone; and (iii) whether a surface is constructed as a surface of revolution/extrusion, a ruled/developable surface, or a generalized cylinder.

Tools from dynamical systems theory are used to decompose 3-D vector fields into simpler building blocks consisting of critical points and phase portraits. A robust critical point detector is developed for three dimensions. Samples from the vector field surrounding each critical point are then used to estimate the associated linear phase portrait, which is written as a 3 × 3 matrix. The estimated matrix may be categorized into one of seven canonical forms by its eigenvalues, which remain consistent under an arbitrary differentiable mapping of the region. The original vector field behavior is estimated using two methods. In one technique, the global behavior is reconstructed using a weighted superposition of phase portraits. For more complex field patterns, a regular partition is imposed prior to phase portrait representation, and each individual partition is decomposed into a separate phase portrait. These methods provide a means of extracting the relevant features and information from the vector field in the form of a higher level descriptor and provide a means of reconstructing the field qualitatively from those descriptors. The method is demonstrated on fluid flow data.

This paper presents a procedure for segmenting planar curves, mainly the projected boundary contours of machined parts, into straight-line segments and elliptical arcs. The break points are divided into two types: corners and smooth joins. The corners are detected by first applying adaptive smoothing to the tangent orientation along the curve, then taking the derivative of the smoothed tangent orientation, and finally locating the high spikes on the derivative. The smooth joins are first roughly located by a dynamic focusing fitting technique and then refined by an adjustment algorithm. The dynamic focusing fitting technique holds one end of a curve segment (which is bounded by a pair of adjacent corners) fixed and scans it from the other end until it focuses on a component segment which fits either a straight line or an elliptical arc. This component segment is identified and the process is repeated in the same manner for the rest of the curve. In the refining stage, each smooth join is adjusted to the left or to the right, point by point, until the measurement of goodness of fit for the curve segment is optimized. Tests of the procedure were performed with the boundary curves of three real object images.

An adaptive early jump-out technique for speeding up the block-based motion estimation is proposed. By using the new technique, we can speed up the full range search several times without losing the picture quality significantly. The proposed technique can also be embedded in almost all the existing fast motion estimation algorithms to speed up the computation further. Since the proposed technique can be embedded in the existing motion estimation algorithms, it can be applied to almost all the standard video codecs, such as the MPEG coder, and improve the coding speed of such codecs significantly. Our technique has been tested on the H.261, the H.263, and the MPEG-I codecs, and the coding speed does improve significantly.