Segmentation is an important step in image analysis for remote sensing; it is responsible for splitting the image into distinct parts. Current segmentation methods used in images of this type are usually based on an a...
详细信息
Segmentation is an important step in image analysis for remote sensing; it is responsible for splitting the image into distinct parts. Current segmentation methods used in images of this type are usually based on an analysis of the pixel values in all image bands. This type of analysis can be computationally expensive. In order to reduce the processing time in the image segmentation process, this paper introduces a parallel approach to image segmentation. This method is based on a region-growing technique. The parallelism is done using the SPMD (single process, multiple data) paradigm and the message-passing model under MPI (Message Passing Interface). The results have been compared with the output of a sequential system executing in the same environment.
The most effective optical method for evaluating donated corneas is to count the number of living endothelial cells. The evaluation of the whole endothelium is usually done in a slit lamp, and the aspects of the cells...
详细信息
The most effective optical method for evaluating donated corneas is to count the number of living endothelial cells. The evaluation of the whole endothelium is usually done in a slit lamp, and the aspects of the cells are subjectively observed. In order to overcome the limitations of subjective observation and to provide an analysis of several parts of the cornea, we have developed a system, attached to the slit lamp, with a dedicated software package for image treatment and cell counting using mathematical morphology. The results are in good agreement with measurements done with the interactive counting method (the correlation coefficient is 0.95798).
This paper presents a multiple point hierarchical approach to the stereo correspondence problem in computer vision. The low-level processing employs an area-and-token hybrid method to obtain, for distinctive points in...
详细信息
This paper presents a multiple point hierarchical approach to the stereo correspondence problem in computer vision. The low-level processing employs an area-and-token hybrid method to obtain, for distinctive points in one image, a set of points in the other image that are candidates for correspondence. The refinement of the set of low-level correspondences obtained is performed by a high-level N-point simultaneous correspondence process, a new constraint introduced for this problem. The high-level processing uses a genetic algorithm approach for searching the solution space. Experimental results show the effectiveness of the method on real world scenes.
Two new Euclidean distance transform algorithms are described. The algorithms are designed using a shortest path graph-search framework. The distance transform can be seen as the solution of a shortest path forest pro...
详细信息
Two new Euclidean distance transform algorithms are described. The algorithms are designed using a shortest path graph-search framework. The distance transform can be seen as the solution of a shortest path forest problem. Previous works have dealt with the Euclidean distance transform (EDT) and with the shortest path forest problem, but none of them have presented an EDT using the graph-search approach. The proposed algorithms are very simple and yet belong to the class of one of the most efficient sequential algorithms. The algorithms easily extend to higher dimensions.
The authors discuss several approaches for providing visualization solutions over the WWW, exploiting available technology for programming Web based applications. An implementation that supports interactive visualizat...
详细信息
The authors discuss several approaches for providing visualization solutions over the WWW, exploiting available technology for programming Web based applications. An implementation that supports interactive visualization of user data through a set of Java applets that interface with the Visualization Toolkit is presented, and its advantages and limitations are discussed. Such an implementation provides a general framework for providing high quality visualization resources on the WWW.
We present current developments towards a real-time method for segmentation of images for robotics applications. Our algorithm follows a fast scan line procedure (rows and columns), thus breaking the 2D segmentation p...
详细信息
We present current developments towards a real-time method for segmentation of images for robotics applications. Our algorithm follows a fast scan line procedure (rows and columns), thus breaking the 2D segmentation problem in two 1D processes. In the first phase we use polynomial functions to find possible 1D boundaries in the image columns and rows separately. In a second step, a fusion method is applied, removing redundancies and false boundaries to find the regions of interest in the segmented images. We try to improve the algorithm using a coarse to fine multi-resolution approach. Preliminary results demonstrate possible applicability in real-time.
A three-dimensional map is a partition of a 3D manifold into topological polyhedra. We consider the problem of visualizing the topology of a three-dimensional map given only its combinatorial description. Our solution...
详细信息
A three-dimensional map is a partition of a 3D manifold into topological polyhedra. We consider the problem of visualizing the topology of a three-dimensional map given only its combinatorial description. Our solution starts by automatically constructing a "nice" geometric realization of the map in R/sup m/, for some m/spl ges/4. The geometric realization is chosen by optimizing certain aesthetic criteria, measured by energy functions. We then project this model to R/sup 3/, and display the resulting multi-celled solid object with a variety of specialized rendering techniques.
We present a novel algorithm for shape estimation, using both brightness and brightness gradient as input data. Our algorithm is an improvement of the recently introduced Green's function approach to shape-from-sh...
详细信息
We present a novel algorithm for shape estimation, using both brightness and brightness gradient as input data. Our algorithm is an improvement of the recently introduced Green's function approach to shape-from-shading (GSFS). In GSFS, we assume that the single brightness image will be matched to a second image through a uniform disparity field, and solve for the matching image via Green's function. When a linear expansion of the reflectance map is considered, the matching image can be related to surface gradient, leading to a closed-form depth map whose free parameters are easily estimated. The author shows that the same procedure can be repeated with the gradient image as input; a second depth estimate thus results which takes into account higher-frequency components of the imaged surface. Extensive experimentation with synthetic and real images corroborates the advantage of the new method.
The paper presents a new approach to image 3D interpolation using Bayesian estimation theory. Two methods are presented, depending on the used a priori information. The approach is applied to the interpolation of X-ra...
详细信息
The paper presents a new approach to image 3D interpolation using Bayesian estimation theory. Two methods are presented, depending on the used a priori information. The approach is applied to the interpolation of X-ray images from the solar atmosphere taken at different depths, as well as RMN tomographic images of a tomato.
The paper presents a framework for distributed reactive characters on a computer network and proposes innovative concepts for shared state management. Firstly, behavioral accuracy is defined in terms of three types of...
详细信息
The paper presents a framework for distributed reactive characters on a computer network and proposes innovative concepts for shared state management. Firstly, behavioral accuracy is defined in terms of three types of state (physical, procedural and emotional). Secondly, visual soundness, a concept inversely proportional to behavioral accuracy, is associated with the autonomy of clones. Thirdly, the usual concept of dead reckoning is relaxed by exploring the idea of autonomy, which is the basis for networked reactive characters. The framework, built on top of an open architecture toolkit called Bamboo, can gracefully cope with reactive environments producing good levels of visual soundness, which results in smooth animations, and behavioral accuracy.
暂无评论