Kenneth A. Byrd

Performance assessment of mammography image segmentation algorithms

In this paper, we present a comprehensive validation analysis to evaluate the performance of three existing mammogram segmentation algorithms against manual segmentation results produced by two expert radiologists. These studies are especially important for the development of computer-aided cancer detection (CAD) systems, which will significantly help improve early detection of breast cancer. Three typical segmentation methods were implemented and applied to 50 malignant mammography images chosen from the University of South Floridas Digital Database for Screening Mammography (DDSM): (a) region growing combined with maximum likelihood modeling (Kinnard model), (b) an active deformable contour model (snake model), and (c) a standard potential field model (standard model). A comprehensive statistical validation protocol was applied to evaluate the computer and expert outlined segmentation results; both sets of results were examined from the inter- and intra-observer points of view. Experimental results are presented and discussed in this communication

The use of spectral skin reflectivity and laser doppler vibrometry data to determine the optimal site and wavelength to collect human vital sign signatures

The carotid artery has been used extensively by researchers to demonstrate that Laser Doppler Vibrometry (LDV) is capable of exploiting vital sign signatures from cooperative human subjects at stando. Research indicates that, the carotid, although good for cooperative and non-traumatic scenarios, is one of the first vital signs to become absent or irregular when a casualty is hemorrhaging and in progress to circulatory (hypovolemic) shock. In an effort to determine the optimal site and wavelength to measure vital signs off human skin, a human subject data collection was executed whereby 14 subjects had their spectral skin reflectivity and vital signs measured at five collection sites (carotid artery, chest, back, right wrist and left wrist). In this paper, we present our findings on using LDV and re ectivity data to determine the optimal collection site and wavelength that should be used to sense pulse signals from quiet and relatively motionless human subjects at stando. In particular, we correlate maximum levels of re ectivity across the ensemble of 14 subjects with vital sign measurements made with an LDV at two ranges, for two scenarios.

Dual IR spectral video inspection of a concealed live animal

Multi-spectral videos have been used in many different fields, however, most prevalently, in the military and medical application areas. Computer vision experts are especially interested in using satellite-grade infrared (IR) sensors for object detection, recognition and identification (DRI) tasks. There has however, been increased interest in using multi-spectral videos for tasks such as inspection/surveillance, image synthesis, N-D object modeling, collision avoidance and intelligent navigation system development. Implicit in the acquisition and processing of videos (and images) for 3D rigid-body objects are the issues of restoration and registration via the traditional affine transformation. In this paper, we present a dynamic scheme for passive-ID recognition of a 3D deformable object, a live hamster; a recognition-ID generated by the fusion of a long-infrared (LIR), 8-12 mum, and middle-infrared (MIR), 5-8 mum, video cameras. This fusion, based on the nonlinear blind demixing of pixels, was previously applied to perform early passive breast cancer detection. By combining blind pixel demixing of a pair of spectral videos (image sequences) and the adaptive neighborhood histogram modification method, we have generalized local video restoration and registration for a live animal in a concealed environment.

Human task performance baseline: results from a cross-band facial identification perception study

Cross-band facial recognition is a difficult task, even for the most robust matching algorithms. Inherent factors such as camera effects (blur, noise, and sampling), and variation in pose and illumination, are known to negatively affect algorithm performance. Because cross-band matching algorithms are in the infancy of development, it is currently unclear if their performance is superior to human observers performing this task. In this paper, we present findings from a pilot study aimed at analyzing the ability of an ensemble of human observers to perform the 1:N cross-band facial identification task on degraded facial images, where the probe and gallery images were captured in different spectral bands (visible, SWIR, MWIR and LWIR). Results from our 11-alternative forced choice perception study indicate that: 1) a group of observers familiar with even a subset of subjects in a gallery set are, on average, able to perform the task with higher probability (p > 0.15) than a group of observers with no prior exposure, and 2) task performance for both the familiar and unfamiliar groups increased 1.5-3.4% when matching multi-spectral probe images to galleries of 24-bit color facial images vs. 8-bit monochrome facial images. For the SWIR case, however, we observed a 9.1% increase in performance with 24-bit facial images vs. 8-bit facial images. Results from this study can be leveraged for future work directly comparing cross-band matching performance of humans vs. algorithms.

Atmospheric turbulence and sensor system effects on biometric algorithm performance

Biometric technologies composed of electro-optical/infrared (EO/IR) sensor systems and advanced matching algorithms are being used in various force protection/security and tactical surveillance applications. To date, most of these sensor systems have been widely used in controlled conditions with varying success (e.g., short range, uniform illumination, cooperative subjects). However the limiting conditions of such systems have yet to be fully studied for long range applications and degraded imaging environments. Biometric technologies used for long range applications will invariably suffer from the effects of atmospheric turbulence degradation. Atmospheric turbulence causes blur, distortion and intensity fluctuations that can severely degrade image quality of electro-optic and thermal imaging systems and, for the case of biometrics technology, translate to poor matching algorithm performance. In this paper, we evaluate the effects of atmospheric turbulence and sensor resolution on biometric matching algorithm performance. We use a subset of the Facial Recognition Technology (FERET) database and a commercial algorithm to analyze facial recognition performance on turbulence degraded facial images. The goal of this work is to understand the feasibility of long-range facial recognition in degraded imaging conditions, and the utility of camera parameter trade studies to enable the design of the next generation biometrics sensor systems.

Verification and validation of a patient simulator for test and evaluation of a laser doppler vibrometer

In the medical community, patient simulators are used to educate and train nurses, medics and doctors in rendering dierent levels of treatment and care to various patient populations. Students have the opportunity to perform real-world medical procedures without putting any patients at risk. A new thrust for the U.S. Army RDECOM CERDEC Night Vision and Electronic Sensors Directorate (NVESD), is the use of remote sensing technologies to detect human vital signs at stando distances. This capability will provide medics with the ability to diagnose while under re in addition to helping them to prioritize the care and evacuation of battleeld casualties. A potential alternative (or precursor) to human subject testing is the use of patient simulators. This substitution (or augmenting) provides a safe and cost eective means to develop, test, and evaluate sensors without putting any human subjects at risk. In this paper, we present a generalized framework that can be used to accredit patient simulator technologies as human simulants for remote physiological monitoring (RPM). Results indicate that we were successful in using a commercial Laser Doppler Vibrometer (LDV) to exploit pulse and respiration signals from a SimMan 3G patient simulator at stando (8 meters).

A subspace learning approach to evaluating the performance of image fusion algorithms

The fusion of multi-spectral images is an important pre-processing operation for scientists and engineers seeking to design robust detection, recognition and identification (DRI) systems. Due to the multitude of pixellevel fusion algorithms available, there is an extreme need for reliable metrics to analyze their performance. Most recently, subspace learning methods have been applied to the field of information fusion for object recognition and classification. This paper aims to extend the capabilities of existing nonlinear dimensionality reduction algorithms to a new area, evaluating the performance of image fusion algorithms. We prove that distances between points in the low dimensional embedding are essentially equivalent to the results given by estimating the amount of information transfered from source images to resultant fused images (normalized mutual information).

Towards non-cooperative standoff biometrics using extremely low frequency (ELF) processing

Earths Field Nuclear Magnetic Resonance (⊕-NMR) is a novel detection technique that exploits the uniformity of the earths magnetic field to measure precessing magnetizations of isotope-having objects such as human beings. The ability to provide standoff medical surveillance will increase the Biomedical Wellness (BMW) of not only aging American citizens but also persons we seek to track from healthy to diseased states. Taking advantage of the uniform field in the earth, spatio-temporal integration (3 orders due to time and 4 orders due to space) and Higher Order Statistics (HOS), namely, change detection via kurtosis (ΔK), SNR and the ability to make non-locality measurements may be increased to at least the level of the typical hospital NMR devices. Moreover, we present theoretical calculations of magnitude and consider time dynamics of the resonant/relaxing magnetic fields as a function of 3D human motion and distance.

Implications of the Advanced Mini-Max (AMM) Classifier on Non-Cooperative Standoff Biometrics

AMM classification is an advanced version of the typical nearest neighbor classifier that allows one to minimize interclass dispersion while at the same time, maximizing intraclass separation. A technique based on the simple orthogonal feature space of Pentland eigenfaces, the combination of these two embodiments will become essential components to a non-cooperative standoff biometric system for military, medical and homeland security applications. The incorporation of robotic assistance further pushes the frontiers of possible surveillance and authentication that can be realized with such a system. The ability to perform out of the line of sight (OLS)-based surveillance adds an additional dimension, and thus novelty, to the already expanding methods to acquire and process environment-specific data.