Showing papers by "Andrew B. Watson published in 1995"

A comparison of Image Quality Models and Metrics Predicting Object Detection

Abstract: Many models and metrics for image quality predict image discriminability, the visibility of the difference between a pair of images. Some image quality applications, such as the quality of imaging radar displays, are concerned with object detection and recognition. Object detection involves looking for one of a large set of object sub-images in a large set of background images and has been approached from this general point of view. We find that discrimination models and metrics can predict the relative detectability of objects in different images, suggesting that these simpler models may be useful in some object detection and recognition applications. Here we compare three alternative measures of image discrimination, a multiple frequency channel model, a single filter model, and RMS error.

Spatial and Spatial Frequency Spreads Of Masking: Measurements and a Contrast-Gain-Control Model

Abstract: Masked pattern detection is affected by the mask's proximity to the target in both the spatial and the spatial frequency domains. We measured contrast thresholds for detection of a Gabor patch centered in a sample of static band-pass noise with a central noise-free aperture. To examine spatial spread, target and mask frequency were identical, at either 2, 4, or 8 cycles/degree (cpd), and the radius of the aperture was 0, 1/2 or 1 cycle of the noise band's center frequency. The results suggest that the spatial spread of masking is scale invariant, with the largest radius producing little masking. To examine spatial frequency spread, a radius of 0 (no aperture) was used with all pairings of 2, 4, and 8 cpd for target and mask. The results suggest that masking is asymmetrical over log frequency: 8 z 0 cpd noise masks a 2 cpd target, but 2 cpd noise does not mask an 8 cpd target. We have used these results to calibrate a model of contrast-gain-control.

Object Detection in Natural Backgrounds Predicted by Discrimination Performance and Models

Abstract: In object detection, an observer looks for an object class member in a set of backgrounds. In discrimination, an observer tries to distinguish two images. Discrimination models predict the probability that an observer detects a difference between two images. We compare object detection and image discrimination with the same stimuli by: (1) making stimulus pairs of the same background with and without the target object and (2) either giving many consecutive trials with the same background (discrimination) or intermixing the stimuli (object detection). Six images of a vehicle in a natural setting were altered to remove the vehicle and mixed with the original image in various proportions. Detection observers rated the images for vehicle presence. Discrimination observers rated the images for any difference from the background image. Estimated detectabilities of the vehicles were found by maximizing the likelihood of a Thurstone category scaling model. The pattern of estimated detectabilities is similar for discrimination and object detection, and is accurately predicted by a Cortex Transform discrimination model. Predictions of a Contrast- Sensitivity- Function filter model and a Root-Mean-Square difference metric based on the digital image values are less accurate. The discrimination detectabilities averaged about twice those of object detection.