Classification Images

The goal of the Classification Images Project is to provide a methodology that allows the identification of the target features that the observers are actually looking for in a detection task. The technological advancement provided by the project is the development for visual psychophysics of a methodology familiar to engineers, the use of pseudo-random noise to identify a transfer function.
The major accomplishments of the project have been:
1) Demonstration of the ability of the method to uncover critical stimulus features in a vernier acuity task and a letter discrimination task by direct visualization of the classification images.
2) Development of data analysis techniques for evaluating the adequacy of proposed models for discrimination when pseudo-random noise has been added to the stimuli.

Presentations and Papers:

A. J. Ahumada, Jr. (1996) Perceptual classification images from Vernier acuity masked by noise. Perception, 26, ECVP Supplement, 18 (Abstract). http://vision.arc.nasa.gov/publications/ecvp96a/abs.html

A. B. Watson, R. Rosenholtz (1997) A Rorschach Test for Visual Classification Strategies. Investigative Ophthalmology and Visual Science, 38, 4, S1, ARVO abstract.

B. L. Beard, A. J. Ahumada, Jr. (1997) Relevant image features for Vernier acuity. Perception, 26, ECVP Supplement, 118 (Abstract).

B. L. Beard, A. J. Ahumada, Jr. (1998) Technique to extract relevant image features for visual tasks. B. E. Rogowitz and T. N. Pappas Human Vision and Electronic Imaging III SPIE Proceedings 3299 Paper 10. Presented at the 1998 IS&T/SPIE Electronic Imaging Symposium, January 24-30, San Jose, CA. http://vision.arc.nasa.gov/personnel/al/papers/98spie/98spie.htm


A. J. Ahumada, Jr., B. L. Beard (1998) Response Classification Images in Vernier Acuity. Investigative Ophthalmology and Visual Science, 39, 4, S1109, ARVO abstract. http://vision.arc.nasa.gov/personnel/al/talks/98arvo/class/abs.html

A. B. Watson (1998) Multi-Category Classification: Template Models and Classification Images. Investigative Ophthalmology and Visual Science, 39, 4, S912, ARVO abstract.

Spatial Vision Models

BACKGROUND

Aviation system designers and evaluators frequently need to know how visible a target, a display element, or an image compression artifact will be over a range of conditions. Sometimes psychophysical measurements can answer this question, but often a computational model is what the designer or system engineer really needs, because the system is not yet realized or the conditions are too numerous.

OBJECTIVES

Provide models that predict target visibility in digital images.
The goal of this program is to develop computational models which take as input computer images or video sequences of such images, and give as output the probability that an observer can see the difference between a pair of such images or sequences or the probability that specific targets are detectable in such images.

APPROACH

Obtain visual detection data from the literature and in the lab, then develop models to predict the data. The model development is guided by biological vision science as well as psychophysical measurements. While the discrimination models are intended to reflect the limitations of sensory processing, target detection involves the selection of target features as a function of the image content. Techniques will be further developed to examine the learning of target features, to identify observer target templates, and to predict detection performance in noisy or cluttered images when these processes become important.

PARTNER

Albert J. Ahumada, Jr., NASA AMES Research Center, Human Information Processing Research Branch

CUSTOMERS

The models developed by this research program will be of utility to designers of displays and image communication systems in aviation, space, medicine, and communications. They are needed for those building computer-aided design tools and those modeling systems to evaluate safety and performance.

METRICS

Errors of prediction, absolute and compared with other available models.
Adoption of models and model subsystems as indicated by references.
Invitations to present models to scientific and technical communities.

NEW MAJOR ACCOMPLISHMENT

Development of easy-to-compute models for image sequence discrimination (Ahumada, Beard & Eriksson, 1998 SPIE).

EXAMPLE APPLICATIONS

Synthetic displays: Evaluation of detectabiility of sensed targets in display fused with database information.
System models: Prediction of probability that pilot will see an out-the-window target that is not transponding.

PROGRAM PLANS (MAJOR MILESTONES)

Complete contrast masking model (FY98) Extend models to include:
Motion (FY99)
Random sensor noise (FY01)

RTOP: 548-50-12