CVGIP: Graphical Models and Image Processing最新文献

Histogram modification can improve the contrast of an image. Histogram equalization has been the most popular histogram modification technique. However, the technique has the tendency to magnify local noise for images with large homogeneous regions. In this paper we suggest a new histogram modification technique which utilizes the intensity distribution of the edge pixels of an image. We first identify the edge pixels of an image. Then the intensity histogram of the edge pixels is constructed. An intensity transformation function is derived from the edge-pixel histogram and then applied to the entire image. In general, this transformation will increase the intensity difference between neighboring homogeneous regions. We also have suggested three tools to measure the performance of contrast-enhancing methods. The three measurements are image contrast value, image information loss value, and local intensity variance value. Our goal for enhancing is to significantly increase an image's contrast value while keeping both the information loss value and the local intensity variance value low. In the experiments, we have compared the performance of the suggested method with that of the ordinary histogram equalization technique and the local area histogram equalization (LAHE) technique using both synthetic and real images. The results were then evaluated by the three tools. The suggested method performed very well both analytically and visually.

The application of the distance transform (DT) to three-dimensional image data, collected, for example, by a confocal scanning laser microscope, requires special modifications to account for the possible anisotropic nature of these images. If anisotropic sampling is not properly accounted for, large errors can occur when simple image processing and image analysis operations related to the DT, such as erosions, dilations, skeletonizations, and distance measurements, are performed. This paper presents a simple, relatively fast way to account for this problem that is suitable for the large sizes typically associated with 3-D images.

This paper presents a constraint method for physically based computer graphics models, based on the constraint stabilization method of Baumgarte and on the dynamic constraints of Barzel and Barr. These new constraints are called generalized dynamic constraints (GDCs). GDCs extend dynamic constraints to obey the principle of virtual work and to fulfill time-varying and inequality constraints. The constraint forces of GDCs are computed by a sparse linear system and are proportional to the Lagrange multipliers of the constraints. GDCs are used to assemble deformable computer graphics models and to simulate collisions between the models.

Multiresolution image processing and analysis has become popular in recent years. One of the most important factors for the success of such systems is the preservation of edges in the process of producing images with reduced resolutions. In this paper 10 image reduction methods are introduced and a comparative evaluation is presented by using a set of synthetic test images and several real images. The quantitative evaluation employs an error measure based on normalized mean-square errors and a set of well-defined image parameters. Edge separation parameter is found to have a strikingly decisive impact on the edge preservation in the context of image reduction. Noise and edge width also show their significant effects. A normalized local intensity variance is studied to bridge the gap between the simple synthetic images and the real images. Finally, suitable methods for producing multiresolution images are recommended.

We introduce a mathematical framework suitable for a general theory of surfaces, objects, and their borders and boundaries in multidimensional discrete spaces. Our motivation comes from practical applications, where objects and their boundaries need to be identified in multidimensional data sets with the further aim of displaying them on a computer screen. Our definitions are biased toward such applications. In particular, we desire to characterize surfaces with a well-determined inside and outside and to define object borders and boundaries so that they will indeed be surfaces of this type. Furthermore, we make our presentation general enough to incorporate many of the reasonable but ad hoc ways that notions of “object,” “border,” and “boundary” may be defined in digital geometry. Some basic theorems are proven, showing that the framework is appropriate for the mathematical treatment of the intuitive notion of a “surface with a connected inside and a connected outside” (a Jordan surface) in the discrete multidimensional environment.

We describe an algorithm for generating connected skeletons of objects in a binary image. The algorithm combines essentially all desirable properties of a skeletonization method: (1) the skeletons it produces have the same simple connectivity as the objects; it is based on a distance transform and can use any “natural” distance metric (in particular those giving a good approximation to the Euclidean distance), resulting in skeletons that are both (2) well-centered and (3) robust with respect to rotation; the skeletons allow the objects to be reconstructed either (4) exactly or (5) approximately to within a specified error; (6) for approximate reconstruction, the skeletons are insensitive to “border noise” without image prefiltering or skeleton postpruning; (7) the skeletons can be thin; (8) the algorithm is fast, taking a fixed number of passes through the image regardless of the width of the objects; and (9) the skeletons have a pleasing visual appearance. Several of these properties may conflict. For example, skeletons cannot always be both thin and allow exact reconstruction and our algorithm can be run to give priority to either property. This paper describes the skeletonization algorithm, discusses the tradeoffs involved and summarizes the formal proofs of its connectivity and reconstructability properties. Because the algorithm is fast, robust, flexible, and provably correct, it is ideally suited for many of the applications of skeletonization—data compression, OCR, shape representation and binary image analysis. The quality of the skeletons produced is demonstrated with numerous examples.

The recognition of objects from imagery in a manner that is independent of scale, position and orientation may be achieved by characterizing an object with a set of extracted invariant features. Several different recognition techniques have been demonstrated that utilize moments to generate such invariant features. These techniques are derived from general moment theory which is widely used throughout statistics and mechanics. In this paper, basic Cartesian moment theory is reviewed and its application to object recognition and image analysis is presented. The geometric properties of low-order moments are discussed along with the definition of several moment-space linear geometric transforms. Finally, significant research in moment-based object recognition is reviewed.

In this paper, an application of Quadrature Mirror Filter (QMF) bank-based subband decomposition to texture analysis is presented. Two-dimensional 4-band QMF structure is used and the QMF features are introduced such that the low-low band extracts the information of spatial dependence and the low-high, high-low, and high-high bands extract the structural information. This approach has the twin advantages of efficient information extraction and parallel implementation. The classification abilities of QMF features are compared to those of Haralick features. The experiments demonstrate that the QMF features have better performance than the Haralick features.

A new approach using the statistical feature matrix, which measures the statistical properties of pixel pairs at several distances, within an image, is proposed for texture analysis. The major properties of this approach are that (1) the size of the matrix is dependent on the maximum distance used instead of the number of gray-levels, (2) the matrix can be expanded easily and (3) some physical properties can be evaluated from the matrix. These properties have enhanced the practical applications of the matrix. In this paper, the matrix is applied to texture classification and visual-perceptual feature extraction. For texture classification, two experiments are performed. First, 16 Brodatz textures are employed to evaluate the performance of the matrix. A simple distance measure is defined to determine the similarity between two statistical feature matrices. Texture discrimination in an additive noise environment is also considered. Second, we apply the matrix to the classification of 150 sampled ultrasonic liver images. From experimental results it can be found that our approach is better than the spatial gray-level dependence method and the spatial frequency-based method. For visual-perceptual feature extraction, we evaluate five basic texture features, namely, coarseness, contrast, regularity, periodicity and roughness, from the statistical feature matrix. It is shown that the statistical feature matrix is an excellent tool for texture analysis.

Reconstruction of the original curve (and the estimation of its parameters) from its digitization is a challenging problem as quantization always causes some loss of information. So we often estimate at least one (or all) continuous curve(s) which is (are) isomorphic to the original one under discretization. Some work has already been done in this respect on straight lines, circles, squares, etc. In this paper, we have attempted this problem for a specialized class of conics which are said to be in normal positions. In normal position the center of the conic is situated at a grid point and its axes are parallel to the coordinate axes. For circles and parabolas, we can directly formulate the domain, i.e., the entire set of continuous curves which produces the same digitization. For ellipses (and this can be extended to hyperbolas too), we first compute the smallest rectangle containing the domain of the given digitization and then estimate the domain itself. The major contribution of this paper lies in the development of a new method of analysis (via the iterative refinement of parameter bounds) which can be easily extended to other 1- or 2-parameter piecewise monotonic shapes such as straight lines or circles with known radius.