Philip A. Araman

A prototype vision system for analyzing CT imagery of hardwood logs

To fully optimize the value of material produced from a hardwood log requires information about type and location of internal defects in the log. This paper describes a prototype vision system that automatically locates and identifies certain classes of defects in hardwood logs. This system uses computer tomograph (CT) imagery. The system uses a number of processing steps. A set of basic features are defined to capture basic 3-D characteristics of wood defects. For 3-D object (defect) recognition, a set of hypothesis tests are employed that use this set of features. To further help cope with the above mentioned variability, the Dempster-Shafer theory of evidential reasoning is used to classify defect objects. Results of preliminary experiments employing two different types of hardwood logs are given.

Active contours on statistical manifolds and texture segmentation

A new approach to active contours on statistical manifolds is presented. The statistical manifolds are 2-dimensional Riemannian manifolds that are statistically defined by maps that transform a parameter domain onto a set of probability density functions. In this novel framework, color or texture features are measured at each image point and their statistical characteristics are estimated. This is different from statistical representation of bounded regions. A modified Kullback-Leibler divergence, that measures dissimilarity between two density distributions, is added to the statistical manifolds so that a geometric interpretation of the manifolds becomes possible. With this framework, we can formulate a metric tensor on the statistical manifolds. Then, a geodesic active contour is evolved with the aid of the metric tensor. We show that the statistical manifold framework provides more robust and accurate texture segmentation results.

Spline curve matching with sparse knot sets: applications to deformable shape detection and recognition

Splines can be used to approximate noisy data with a few control points. This paper presents a new curve matching method for deformable shapes using two-dimensional splines. In contrast to the residual error criterion [F.S. Cohen et al., 1992], which is based on relative locations of corresponding knot points such that is reliable primarily for dense point sets, we use deformation energy of thin-plate-spline mapping between sparse knot points and normalized local curvature information. This method has been tested successfully for the detection and recognition of deformable shapes.

A development system for creating real-time machine vision hardware using field programmable gate arrays

In this paper, we introduce a new development system for creating real-time image processing hardware using custom computing machines with multiple Field Programmable Gate Array (FPGA) chips. Three distinct processes are accomplished within the development system: design entry, verification, and translation. A library of modules that implement common low-level machine vision functions is used to create complex designs based on a dataflow graph representation. The librarys low-level image processing modules contain both gate-level and chip-level hardware components, of which the gate-level components are compiled into the functionality of available FPGA chips. Standard interfaces are established for input/output of the modules, allowing for the creation of sophisticated software support tools. Experimental results verify the utility of this development system for easily creating real-time machine vision hardware using multiple FPGA-based custom computing machines.