A. Lynn Abbott

Optimization of color conversion for face recognition

This paper concerns the conversion of color images to monochromatic form for the purpose of human face recognition. Many face recognition systems operate using monochromatic information alone even when color images are available. In such cases, simple color transformations are commonly used that are not optimal for the face recognition task. We present a framework for selecting the transformation from face imagery using one of three methods: Karhunen-Loève analysis, linear regression of color distribution, and a genetic algorithm. Experimental results are presented for both the well-known eigenface method and for extraction of Gabor-based face features to demonstrate the potential for improved overall system performance. Using a database of 280 images, our experiments using these methods resulted in performance improvements of approximately 4% to 14%.

UAV-Based Stereo Vision for Rapid Aerial Terrain Mapping

Aerial terrain mapping has been used for many years to monitor natural habitats and ecosystems, assist in urban planning, and monitor trends in land usage. Recent improvements in digital imaging, LiDAR, and synthetic aperture radar have facilitated the generation of 3-D terrain models for analysis in these applications. Unfortunately, these systems typically require large manned aircraft and significant post-processing of data before viewable results are produced. This inhibits use of these technologies in time-critical applications such as disaster relief, autonomous obstacle avoidance, and landing-zone assessment for a vertical take-off and landing aircraft. This paper describes a wide-baseline stereo vision system that enables near-real-time generation of dense 3-D terrain maps. The key advantage of computational stereo vision over monocular structure-from-motion is that terrain can be reconstructed from a single synchronized pair of calibrated images. The paper describes a working prototype, and presents a novel approach for combining separate stereo maps into larger terrain mosaics. The new stereo system and algorithm have an accuracy ranging from 56 cm to 65 cm across the field of view at an altitude of 40 m. Also, dense correlation of the imagery generates over 2200 points/m2. The system weighs just 3.1 kg, roughly one-fourth the weight of comparable high-altitude mapping systems, at ca. one-tenth the cost. The paper also describes potential implementations using Field-Programmable Gate Arrays (FPGAs) and Application-Specific Integrated Circuits (ASICs) for real-time operation.

Rule-driven defect detection in CT images of hardwood logs

This paper deals with automated detection and identification of internal defects in hardwood logs using computed tomography (CT) images. We have developed a system that employs artificial neural networks (ANNs) to perform tentative classification of logs on a pixel-by-pixel basis. This approach achieves a high level of classification accuracy for several hardwood species (northern red oak, Quercus rubra, L., water oak, Q. nigra, L., yellow poplar, Liriodendron tulipifera, L., and black cherry, Prunus serotina, Ehrh.), and three common defect types (knots, splits, and decay). Although the results are very satisfactory statistically, a subjective examination reveals situations that could be refined in a subsequent post-processing step. We are currently developing a rule-based, contextual approach to region refinement that augments the initial emphasis on local information. The resulting rules are domain dependent, utilizing information that depends on region shape and type of defect. For example splits tend to be long and narrow, and this knowledge can be used to merge smaller, disjoint regions that have tentatively been labeled as splits. Similarly, image regions that represent knots, decay, and clear wood can be refined by removing small, spurious regions and by smoothing the boundaries of these regions. Mathematical morphology operators can be used for most of these tasks. This paper provides details concerning the domain-dependent rules by which morphology operators are chosen, and for merging results from different operations.

Machine vision using artificial neural networks with local 3D neighborhoods

Abstract Several approaches have been reported previously to identify internal log defects automatically using computed tomography (CT) imagery. Most of these have been feasibility efforts and consequently have had several limitations: (1) reports of classification accuracy are largely subjective, not statistical; (2) there has been no attempt to achieve real-time operation; and (3) texture information has not been used for image segmentation, but has been limited to region labeling. Neural network classifiers based on local neighborhoods have the potential to greatly increase computational speed, can be implemented to incorporate textural features during segmentation, and can provide an objective assessment of classification performance. This paper describes a method in which a multilayer feed-forward network is used to perform pixel-by-pixel defect classification. After initial thresholding to separate wood from background and internal voids, the classifier labels each pixel of a CT slice using histogram-normalized values of pixels in a 3 × 3 × 3 window about the classified pixel. A post-processing step then removes some spurious pixel misclassifications. Our approach is able to identify bark, knots, decay, splits, and clear wood on CT images from several species of hardwoods. By using normalized pixel values as inputs to the classifier, the neural network is able to formulate and apply aggregate features, such as average and standard deviation, as well as texture-related features. With appropriate hardware, the method can operate in real time. This approach to machine vision also has implications for the analysis of 2D gray-scale images or 3D RGB images.

Implementation of a 2-D Fast Fourier Transform on an FPGA-Based Custom Computing Machine

The two dimensional fast Fourier transform (2-D FFT) is an indispensable operation in many digital signal processing applications but yet is deemed computationally expensive when performed on a conventional general purpose processors. This paper presents the implementation and performance figures for the Fourier transform on a FPGA-based custom computer. The computation of a 2-D FFT requires O(N2log2N) floating point arithmetic operations for an NxN image. By implementing the FFT algorithm on a custom computing machine (CCM) called Splash-2, a computation speed of 180 Mflops and a speed-up of 23 times over a Sparc-10 workstation is achieved.

Image Processing on a Custom Computing Platform

Custom computing platforms are emerging as a class of computing engine that not only can provide near application-specific computational performance, but also can be configured to accommodate a wide variety of tasks. Due to vast computational needs, image processing computing platforms are traditionally constructed either by using costly application-specific hardware to support real-time image processing, or by sacrificing real-time performance and using a general-purpose engine. The Splash-2 custom computing platform is a general-purpose platform not designed specifically for image processing, yet it can cost-effectively deliver real-time performance on a wide variety of image applications. This paper describes an image processing system based on the Splash-2 custom computing engine, along with performance results from a variety of image processing tasks extracted from a working laboratory system. The application design process used for these image processing tasks is also examined.

NONDESTRUCTIVE EVALUATION OF HARDWOOD LOGS: CT SCANNING, MACHINE VISION AND DATA UTILIZATION

Sawing of hardwood logs still relies on relatively simple technologies that, in spite of their lack of sophisticatio n, have been successful for many years due to wood’s traditional low cost and ready availability. These characteristics of the hardwood resource have changed dramatically over the past 20 years, however, forcing wood processors to become more efficient in their operations. In spite of some recent advances, the breakdown of hardwood logs into lumber continues to be hampered by the inability of sawyers to “see” inside of the log prior to making irreversible cutting decisions. The need for noninvasive assessment of hardwood logs prior to breakdown is well accepted, but is difficult to realize because industrial scanning. in this context, is unique in several respects. For example, large volumes of material must be inspected quickly over an extended duty cycle, the wood material still possesses relatively low value compared to other industrial materials that require internal scanning, and many wood processors are small operations located in rural areas. Successful implementation of new scanning technology, however, will have tremendous payback for wood processors. and for timber resource conservation efforts. The research program reviewed here applies a three-pronged approach to address this situation. First, a relatively new and innovative CT scanning technology is being developed that can scan hardwood logs at industrial speeds. Second. machine vision software has been created that can interpret scanned images rapidly and with high accuracy. Third, we have developed 3-D rendering and analysis techniques that will enable sawyers to apply image assessment to actual log

Stereo Analysis for Vision-based Guidance and Control of Aircraft Landing

This work considers the feasibility of binocular stereo imaging for use in aircraft guidance and control. Of particular interest is the use of stereo-based ranging as an aid in autonomous or semiautonomous landing of General Aviation (GA) class aircraft. With monocular image analysis, an aircraft’s position can be estimated up to a scale factor only, unless prior knowledge (e.g., width of the runway) is available. With calibrated stereo cameras, complete three-dimensional position and orientation estimates are possible when point correspondences are known for the two images. The major difficulty, however, lies in the development of algorithms that can identify the runway and determine point correspondences reliably. This paper presents a method for detecting runway edges and estimating distances from the aircraft to the touchdown point during an approach. The method has been tested using stereo image sequences obtained during flight tests in the Hampton Roads, Virginia area. These tests have allowed us to examine the effectiveness of stereo matching techniques for images of precision and nonprecision runways during landing operations. Additional tests involving feature tracking and optic flow recovery were conducted at a visual runway. Although full calibration was not available for accuracy assessment, regression analysis indicated very good results (

Auto-Optimized Multimodal Expression Recognition Framework Using 3D Kinect Data for ASD Therapeutic Aid

This paper concerns the automatic recognition of human facial expressions using a fast 3D sensor, such as the Kinect. Facial expressions represent a rich source of information regarding emotion and interpersonal communication. The ability to recognize expressions automatically will have a large impact in many areas, particularly human-computer interaction. This paper describes 2 frameworks for recognizing 6 basic expressions using 3-dimensional data sequences that are captured in real time. Results are presented that demonstrate accuracy levels for the different techniques, and for different methods of preprocessing, registration and classification. We also describe the potential to use such a system for treatment of children with autism spectrum disorders (ASD).

High-Speed Region Detection and Labeling Using an FPGA Based Custom Computing Platform

General purpose custom computing platforms, such as Splash-2, have demonstrated the ability to enter mainstream computing not only due to their near application-specific speeds but also because of their ability to run a wide variety of tasks. Splash-2 is a second-generation FPGA-based system that can deliver processing performance rivaling application-specific systems, but is also reconfigurable. This paper describes a computationally intensive image processing task, known as region labeling, which demonstrates the effectiveness of such platforms. The design and implementation of a region labeling task on the Splash-2 custom computing platform are described and the resulting performance is compared with that of other machines.