CVGIP: Graphical Models and Image Processing最新文献

Thresholding is a common image processing operation applied to gray-scale images to obtain binary or multilevel images. Traditionally, one of two approaches is used: global or locally adaptive processing. However, each of these approaches has a disadvantage: the global approach neglects local information, and the locally adaptive approach neglects global information. A thresholding method is described here that is global in approach, but uses a measure of local information, namely connectivity. Thresholds are found at the intensity levels that best preserve the connectivity of regions within the image. Thus, this method has advantages of both global and locally adaptive approaches. This method is applied here to document images. Experimental comparisons against other thresholding methods show that the connectivity-preserving method yields much improved results. On binary images, this method has been shown to improve subsequent OCR recognition rates from about 95% to 97,5%. More importantly, the new method has been shown to reduce the number of binarization failures (where text is so poorly binarized as to be totally unrecognizable by a commercial OCR system) from 33% to 6% on difficult images. For multilevel document images, as well, the results show similar improvement.

We have shown (J. Appl. Phys., 1990, 1415-1420) that deconvolving an image which was blurred by a Gaussian filter is equivalent to antidiffusing the image for an appropriate duration of time. However, the antidiffusion algorithm used to show this, based on backward integration of the diffusion equation, is extremely sensitive to noise with numerical errors increasing exponentially with time. Thus, an extremely high signal to noise ratio is required for reconstruction of a blurred image via antidiffusion. In this paper, we introduce a new antidiffusion algorithm which is substantially more robust with respect to noise. This is because each functional component in the series of the reconstructed image is obtained analytically from a corresponding component of the blurred image. We show that the algorithm yields accurate reconstructions of Gaussian-smeared signals and images with extremely low signal to noise ratios.

A method is presented for detecting blurred edges in images and for estimating the following edge parameters: position, orientation, amplitude, mean value, and edge slope. The method is based on a local image decomposition technique called a polynomial transform. The information that is made explicit by the polynomial transform is well suited to detect image features, such as edges, and to estimate feature parameters. By using the relationship between the polynomial coefficients of a blurred feature and those of the a priori assumed (unblurred) feature in the scene, the parameters of the blurred feature can be estimated. The performance of the proposed edge parameter estimation method in the presence of image noise has been analyzed. An algorithm is presented for estimating the spread of a position-invariant Gaussian blurring kernel, using estimates at different edge locations over the image. First a single-scale algorithm is developed in which one polynomial transform is used. A critical parameter of the single-scale algorithm is the window size, which has to be chosen a priori. Since the reliability of the estimate for the spread of the blurring kernel depends on the ratio of this spread to the window size, it is difficult to choose a window of appropriate size a priori. The problem is overcome by a multiscale blur estimation algorithm where several polynomial transforms at different scales are applied, and the appropriate scale for analysis is chosen a posteriori. By applying the blur estimation algorithm to natural and synthetic images with different amounts of blur and noise, it is shown that the algorithm gives reliable estimates for the spread of the blurring kernel even at low signal-to-noise ratios.

We compare two classes of techniques, cross-covariance-based and Fourier-based, for estimating band-to-band misregistrations in multispectral imagery. We show that both methods often give biased estimates of the misregistrations, the former because of inadequate interpolation procedures and the latter because they do not account for the presence of aliasing. Such aliasing is often present, especially in remote sensing imagery. We describe a Fourier-based method that accounts for aliasing and that, for a variety of 512 × 512 image pairs, gives misregistration estimates with standard errors quite often less than 1/100th of a pixel in both horizontal and vertical directions. The theory is applied to one artificial and three real image pairs, thus demonstrating some of its practical consequences. There is also a brief discussion of the implications of the theory for image registration.

In this paper, we present an efficient three-dimensional (3-D) parallel thinning algorithm for extracting both the medial surfaces and the medial axes of a 3-D object (given as a 3-D binary image). A new Euler table is derived to ensure the invariance of the Euler characteristic of the object, during thinning. An octree data structure of 3 × 3 × 3 lattice points is built to examine the local connectivity. The sets of "simple" points found by different researchers are compared with the constructed set. Different definitions of "surface" points including ours are given. By preserving the topological and the geometrical conditions, our algorithm produces desirable skeletons and performs better than others in terms of noise sensitivity and speed. Pre- and postprocessors can be used to remove additional noise spurs. Its use in defect analysis of objects produced by casting and forging is discussed.

This paper describes a new method for multilevel threshold selection of gray level images. The proposed method includes three main stages. First, a hill-clustering technique is applied to the image histogram in order to approximately determine the peak locations of the histogram. Then, the histogram segments between the peaks are approximated by rational functions using a linear minimax approximation algorithm. Finally, the application of the one-dimensional Golden search minimization algorithm gives the global minimum of each rational function, which corresponds to a multilevel threshold value. Experimental results for histograms with two or more peaks are presented.

The entropy method for image thresholding suggested by Kapur et al. has been modified and a more pertinent information measure of the image is obtained. Essentially this consists of viewing the image as a compositum of two fuzzy sets corresponding to the two classes with membership coefficient associated with each gray level a function of its frequency of occurrence as well as its distance from the intermediate threshold selected. An extension of this technique to consider the semantic content of the image is also discussed. The superiority of the suggested method over artificial histograms modelled by Gaussian distributions is demonstrated. Experimental results on several images are also presented to support the validity of the concepts used.