Julie A. Skipper

Object recognition via information-theoretic measures/metrics

Discrimination of friendly or hostile objects is investigated using information-theory measures/metric in an image which has been compromised by a number of factors. In aerial military images, objects with different orientations can be reasonably approximated by a single identification signature consisting of the average histogram of the object under rotations. Three different information-theoretic measures/metrics are studied as possible criteria to help classify the objects. The first measure is the standard mutual information (MI) between the sampled object and the library object signatures. A second measure is based on information efficiency, which differs from MI. Finally an information distance metric is employed which determines the distance, in an information sense, between the sampled object and the library object. It is shown that the three (parsimonious) information-theoretic variables introduced here form an independent basis in the sense that any variable in the information channel can be uniquely expressed in terms of the three parameters introduced here. The methodology discussed is tested on a sample set of standardized images to evaluate their efficacy. A performance standardization methodology is presented which is based on manipulation of contrast, brightness, and size attributes of the sample objects of interest.

Thermal imaging to detect physiological indicators of stress in humans

Real-time, stand-off sensing of human subjects to detect emotional state would be valuable in many defense, security and medical scenarios. We are developing a multimodal sensor platform that incorporates high-resolution electro-optical and mid-wave infrared (MWIR) cameras and a millimeter-wave radar system to identify individuals who are psychologically stressed. Recent experiments have aimed to: 1) assess responses to physical versus psychological stressors; 2) examine the impact of topical skin products on thermal signatures; and 3) evaluate the fidelity of vital signs extracted from thermal imagery and radar signatures. Registered image and sensor data were collected as subjects (n=32) performed mental and physical tasks. In each image, the face was segmented into 29 non-overlapping segments based on fiducial points automatically output by our facial feature tracker. Image features were defined that facilitated discrimination between psychological and physical stress states. To test the ability to intentionally mask thermal responses indicative of anxiety or fear, subjects applied one of four topical skin products to one half of their face before performing tasks. Finally, we evaluated the performance of two non-contact techniques to detect respiration and heart rate: chest displacement extracted from the radar signal and temperature fluctuations at the nose tip and regions near superficial arteries to detect respiration and heart rates, respectively, extracted from the MWIR imagery. Our results are very satisfactory: classification of physical versus psychological stressors is repeatedly greater than 90%, thermal masking was almost always ineffective, and accurate heart and respiration rates are detectable in both thermal and radar signatures.

Power spectrum weighted edge analysis for straight edge detection in images

Most man-made objects provide characteristic straight line edges and, therefore, edge extraction is a commonly used target detection tool. However, noisy images often yield broken edges that lead to missed detections, and extraneous edges that may contribute to false target detections. We present a sliding-block approach for target detection using weighted power spectral analysis. In general, straight line edges appearing at a given frequency are represented as a peak in the Fourier domain at a radius corresponding to that frequency, and a direction corresponding to the orientation of the edges in the spatial domain. Knowing the edge width and spacing between the edges, a band-pass filter is designed to extract the Fourier peaks corresponding to the target edges and suppress image noise. These peaks are then detected by amplitude thresholding. The frequency band width and the subsequent spatial filter mask size are variable parameters to facilitate detection of target objects of different sizes under known imaging geometries. Many military objects, such as trucks, tanks and missile launchers, produce definite signatures with parallel lines and the algorithm proves to be ideal for detecting such objects. Moreover, shadow-casting objects generally provide sharp edges and are readily detected. The block operation procedure offers advantages of significant reduction in noise influence, improved edge detection, faster processing speed and versatility to detect diverse objects of different sizes in the image. With Scud missile launcher replicas as target objects, the method has been successfully tested on terrain board test images under different backgrounds, illumination and imaging geometries with cameras of differing spatial resolution and bit-depth.

Entropy Selective Mutual Information-Based Image Registration

Multimodal imaging systems demand sophisticated registration routines. Due to computation time and non-convergence issues, the use of traditional mutual information (MI)-based registration is impractical. We propose our sampling optimization technique with selective high entropy MI computation as a rapid and robust image registration method for real-time applications.

Deblurring of X-Ray Spectra Acquired with a Nal-Photomultiplier Detector by Constrained Least-Squares Deconvolution

A constrained least-squares technique to correct diagnostic x-ray tube energy spectra for inherent blurring by scintillation detectors was developed. The measured detector response function to monoenergetic sources was used to construct a matrix that modeled the energy broadening in the crystal. This blurring operator, along with an estimate of statistical noise in the count data, comprised the a priori system knowledge required for application of the method. Tungsten anode spectra up to 90 kVp were acquired with a Nal-photomultiplier detector system at a source-to-detector distance of 30 cm. X-ray tube output was collimated at the detector by a 0.5 mm diameter pinhole collimator. Measured Nal spectra were compared to both published reference data and to spectra acquired in our laboratory with a Ge detector system. Application of the constrained least-squares technique involved first defining a criterion function that combined an assessment of the goodness of fit with a weighted measure of the smoothness of the solution. Minimization of this function resulted in the corrected spectrum. While it is not possible to recover the characteristic tungsten peaks, the success of our method in deconvolving the measured spectra was demonstrated by a significant improvement in agreement with reference data. To provide a measure of this agreement, a histogram of the differences between the two curves was generated. The full width at half maximum (FWHM) of the Gaussian distribution fit to the histogram was used to quantify the similarity between the spectra and the reference data, both before and after correction. As spectral agreement improves, the FWHM becomes smaller. We show that application of the constrained least-squares technique improved spectral matching by 20%-60%.

Human Thermal Modeling to Augment MWIR Image Analysis in Surveillance Applications

The interpretation of thermal imagery can be augmented with information derived from human thermal modeling to better infer human activity during, or prior to, data capture. This additional insight into human activity could prove useful in security and surveillance applications. We have implemented Tanabe’s 65 NM thermocomfort model to predict skin surface temperature under a wide variety of environmental, activity and body parameters. Here, humans are modeled as sixteen segments (head, chest, upper leg, etc.), wherein spherical geometry is assumed for the head and cylindrical geometry is assumed for all other segments. Each segment is comprised of four layers: core, muscle, fat, and skin. Clothing is modeled as an additional layer (or layers) of resistance. Users supply input parameters via our custom MATLAB graphical user interface that includes a robust clothing database based on McCullough’s A Database for Determining the Evaporative Resistance of Clothing, and then Tanabe’s bioheat equations are solved to predict skin temperatures of each body segment. As an initial step of model validation, we compared our computed thermal resistances with literature values. Our evaporative and dry resistance on a per segment basis agreed with literature values. The dry resistance of each segment varied no more than .03 [m2°C/W]. Model validation will be extended to compare the results of our human subject trials (known body parameters, clothing, environmental factors and activity levels) to model outputs. Agreement would further substantiate the propagation of model- predicted skin temperatures through the thermal imager’s transfer function to predict human heat signatures in thermal imagery.

Human Perceptions of Nonverbal Behavior Presented Using Synthetic Humans

Synthetic humans are computer-generated characters that are designed to behave like humans for the purpose of training or entertainment. The purpose of this study was to evaluate the perceptions of subjects interacting with synthetic humans to determine their responses to nonverbal behaviors, realism, and character personality. This study was part of a research program to develop a virtual game to train awareness of nonverbal communication for cross-cultural competency (3C).Three synthetic humans were created with different levels of realism with respect to their facial movements and skin textures. Low realism characters were defined as models purchased from the company Evolver, with additional facial action units (FAU) added to the character’s face. High realism characters were created based on a model of a real person’s head using 3D imaging cameras and a digital video camera. The same FAUs available in the Evolver characters were also coded into the high realism character as well as more realistic skin texture. During a virtual scenario the subject was asked to interview three characters in the U.S. Army. The subject interviewed each character one-on-one. The three computer characters included two white males, and one black female.The results of this study showed that it is possible to create synthetic humans that include nonverbal behaviors and personalities that are perceived by subjects, and that the subject’s own personal lens affected how they perceive the character. For example, the character Brent was rated similarly by most subjects with respect to personality traits as defined by the Big Five Factor Model. However, half the subjects indicated they liked him (friendly and confident), while about half the subjects did not like him (too confident as to be arrogant).Copyright

Target recognition: fusing long-wave infrared and electro-optical imagery for detection of humans in a scene

We aim to identify humans in multimodal imagery by predicting the human long-wave infrared (LWIR) signature in a variety of scenarios. By adapting Tanabes thermocomfort model, we simulate human body heat flow both between tissue layers (core, muscle, fat and skin) and between body segments (head, chest, upper arm, etc.). To assess the validity of our implementation, we simulated the conditions described in actual human subject studies, and compared our results to values reported in the literature. Inputs to the model include age, height, weight, clothing, physical activity and ambient conditions, including temperature, humidity and wind velocity. Iteration of heat transport equations and a thermoregulatory component yields temporal data of segment surface temperature. Our model was found to be in close agreement with experimentally collected data, with a maximum deviation from literature values of approximately 0.80%. By comparing the predicted human thermal signature to deblurred LWIR images and then fusing this information at the feature level with high-resolution electro-optical image data, we can facilitate identity detection of objects in a scene acquired under different conditions. Ultimately, our goal is to differentiate humans from their surroundings and label non-human objects as thermal clutter.

Feasibility of Osteoporosis Screening by Dual-Energy Radiographic Absorptiometry of the Phalanx

A new dual-energy radiographic absorptiometry-based technique is presented as a cost-effective method for mass osteoporosis screening. Designed for use in a dental health care setting, we propose a method and device for bone mineral density (BMD) assessment using the middle phalanx of the 3rd digit as our anatomical measurement site. Our 2-staged project includes the development of the prototype to carry out the measurement and the execution of a small pilot study to determine the efficacy of the method and device. Fifty subjects from the general adult population (age range: 25-82 yr), wherein 10 normal subjects (5 females and 5 males) and 40 target group subjects (30 females and 10 males) who were at risk for osteoporosis (as assessed qualitatively through questionnaire responses) were evaluated with our method. The BMD values obtained from the normal and target groups were significantly different (p<0.0001). Phantom measurements to determine the bias and coefficient of variation of the technique yielded values of 1.9% and 7%, respectively. The proposed technique could provide a relatively inexpensive and widely available means for mass osteoporosis screening. Further validation of this method, to include comparison to a gold standard, such as dual-energy X-ray absorptiometry, is warranted.

Studies on Image Fusion Techniques for Dynamic Applications

A survey of present methods and current techniques being pursued by the US Air Force for image fusion and registration is conducted. Formulating the problem within a signal detection theory framework provides a unique thrust.