Dragana Brzakovic

Camera geometries for image matching in 3-D machine vision

The location of a scene element can be determined from the disparity of two of its depicted entities (each in a different image). Prior to establishing disparity, however, the correspondence problem must be solved. It is shown that for the axial-motion stereo camera model the probability of determining unambiguous correspondence assignments is significantly greater than that for other stereo camera models. However, the mere geometry of the stereo camera system does not provide sufficient information for uniquely identifying correct correspondences. Therefore, additional constraints derived from justifiable assumptions about the scene domain and from the scene radiance model are utilized to reduce the number of potential matches. The measure for establishing the correct correspondence is shown to be a function of the geometrical constraints, scene constraints, and scene radiance model. >

Weld pool edge detection for automated control of welding

A vision system that determines the edges of the weld pool in sequences of gas-tungsten-arc welding images acquired by a coaxial viewing system is described. The vision system uses a transformation that maps the edge of a weld pool into a vertical line. The weld pool edge is detected in the transform domain by using a directional filter, which retains only intensity changes of interest, and a one-dimensional edge detector. The edge of the weld pool, in the physical domain, is determined using the inverse transformation. The transformation uses parameters that are updated when processing a sequence of images, and are initially determined by analyzing the first image frame in the physical domain. >

An approach to defect detection in materials characterized by complex textures

Abstract This paper describes an expert system which detects and categorizes defects in digitized images of materials characterized by complex texture patterns. The system performs two major tasks: image segmentation and defect classification. Segmentation is done in two steps. First, individual texture samples are isolated. Second, a pyramid linking scheme is used to locate defects in each sample. This pyramid linking scheme can be fine-tuned to various textures and to the size of the defects to be detected. The defect classification method is dependent on the material being analyzed. This paper describes a hierarchical defect classification scheme for analyzing wood. The hierarchy incorporates a shape descriptor, size, and textural descriptors of individual defects. The system has been tested on parquet samples where it successfully detects and classifies defects.

Automated iterative noise filtering (image processing)

An iterative method for removing signal-independent, additive noise from digital images is described. The underlying concept of this fully automated method is noise filtering by use of local statistics. Assuming that the noise is statistically stationary, noise variance is estimated from an input image by utilizing its smallest local variances. Noise filtering is performed iteratively and terminates when the estimated noise variance converges to zero. The method successfully processes degraded images by filtering noise from regions of uniform intensity while preserving texture pixels and edges. >

Two methods of image extension

Abstract Most operations in image processing involve application of a mask centered at each pixel in an image. Consequently, in order to preserve image size under various operations, it is necessary to extend an image outside its borders for the number of columns and rows equal to half the mask size. This paper describes two methods of image extension so that image size may be preserved under various image processing operations. An algorithm is provided for each method and the results obtained are compared in terms of their statistical properties and aesthetic value.

Rule-based multitemplate edge detector

Abstract This paper describes an edge detection algorithm that employs directional templates of various sizes and rule-based decision making to assign confidence values to candidate edge pixels. High confidence values indicate a high likelihood of pixels being edge pixels, while low confidence values are associated with pixels likely to be noise. Consequently, an edge image is generated by retaining pixels associated with high confidence values. The performance of the algorithm is quantitatively evaluated on synthetic images. The evaluation concentrates on performance in the presence of noise.

An approach to computer-aided document examination

In this paper we describe a knowledge-based expert system capable of investigating a typewritten document and acquiring evidence on its falsification. A document is assumed to be falsified if its typing has been performed in multiple typing sessions. Verification of document authenticity is carried out by subjecting a document to four tests: alignment analysis, symbol comparison, gray level intensity analysis and type font identification. Tests are performed in succession and failure of any one of the tests signifies possible document falsificaton.

Defect characterization and classification for web materials

The authors describe a development environment for the design and testing of algorithms for automated defect classification in web materials. The environment includes four subsystems: detection, characterization, feature analysis, and classification. A subsystem consists of a number of algorithms from which a specific algorithm is chosen based on the classification objectives. The subsystems are interconnected using direct and feedback loops. The feedback loops are used when evaluating the performance of algorithms in previous subsystems. An example of defect classification is given, including discussion of the results and specific algorithms used by the characterization, feature analysis, and classification subsystems.<<ETX>>

Road edge detection for mobile robot navigation

The authors describe a novel algorithm for the detection of road edges in image sequences for mobile robot navigation. The essence of this algorithm is road modeling by spline functions and edge detection by a linear recursive filter. The filter uses a measurement procedure that involves an edge point search about a specified reference curve. This curve is determined automatically by preprocessing the first frame. Estimates of the road edges in each frame are used as the reference curve for the following frame. The algorithm processes a sequence of road images, and it does not require a priori knowledge about the image contents. The algorithm has been tested on image sequences, yielding satisfactory results.<<ETX>>

Correspondence In Pairs Of Images Acquired By Camera Displacement In Depth

In this study, the correspondence problem is analyzed for two different camera-arrangement models based on their geometrical and topological properties. Emphasis is given to the model where the two images (image planes), acquired by moving a camera along its optical axis, are parallel to one another. The advantage of this arrangement over the one where the two images are coplanar is demonstrated, and a correspondence process regarding the motion-in-depth model is proposed.