Object classification for human and ideal observers.

Z Liu; D C Knill; D Kersten

doi:10.1016/0042-6989(94)00150-k

Home/Archive/PubMed

PubMed · 7900295

Object classification for human and ideal observers.

Z Liu·D C Knill·D Kersten

View original source ↗Export metadata ↓

Abstract

We describe a novel approach, based on ideal observer analysis, for measuring the ability of human observers to use image information for 3D object perception. We compute the statistical efficiency of subjects relative to an ideal observer for a 3D object classification task. After training to 11 different views of a randomly shaped thick wire object, subjects were asked which of a pair of noisy views of the object best matched the learned object. Efficiency relative to the actual information in the stimuli can be as high as 20%. Increases in object regularity (e.g. symmetry) lead to increases in the efficiency with which novel views of an object could be classified. Furthermore, such increases in regularity also lead to decreases in the effect of viewpoint on classification efficiency. Human statistical efficiencies relative to a 2D ideal observer exceeded 100%, thereby excluding all models which are sub-optimal relative to the 2D ideal.

Explore related subjects

Depth Perception Discrimination, Psychological Form Perception Humans Learning Models, Neurological Probability Learning Psychophysics Sensory Thresholds Journal Article Research Support, Non-U.S. Gov't Research Support, U.S. Gov't, Non-P.H.S.

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Z Liu, D C Knill, D Kersten. 1995. Object classification for human and ideal observers.. https://doi.org/10.1016/0042-6989(94)00150-k

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related citations

PubMed1996

Ideal observer for heading judgments.

Several aspects of the viewing situation affect the ability to determine heading from optical flow. These include the amount of depth variation and number of texture elements in the scene, the location and amount of the visual field stimulated, and the position of the focus of expansion within the stimulus. Without a quantification of the discrimination information provided by the stimuli presented to the observer, it is impossible to determine how much of an observed change in performance reflects the properties of neural mechanisms and strategies employed by the observer. To enable a better quantification, we developed an ideal observer for the discrimination of heading from random-dot flow fields. Internal noises of the ideal observer were set by the results of single-dot velocity discrimination experiments. We compared human and ideal observer performance in discriminating headings with different patterns of flow (e.g. radial vs laminar) presented on different parts of the retina. Efficiency--the ratio of ideal and human thresholds--was fairly constant for the various flow patterns and retinal eccentricities. This outcome indicates that most of the variation in human observers' ability to estimate heading from the flow patterns and retinal loci considered here is due to changes in the discrimination information provided by the stimulus after measurement by the visual system. In the discussion, we show how the ideal observer can be used to quantify the spatial distribution of heading discrimination information for any observer translation through any scene represented by dots.

Depth Perception↗

PubMed1996

Spatio-temporal requirements for binocular correlation in stereopsis.

We measured sensitivity to binocular correlation in dynamic random-dot stereograms that defined moving sinusoidal gratings-in-depth. At a range of spatial frequencies and drift rates we established sensitivity by adding Gaussian distributed disparity noise to the modulation of disparity that defined a cyclopean grating, and finding the noise amplitude that rendered the grating just detectable. This permitted correlation thresholds to be measured at a range of suprathreshold disparity amplitudes. Spatial requirements for binocular correlation depend little on temporal frequency, and vice versa. This suggests that binocular correlation mechanisms can be characterized by independent spatial and temporal sensitivity functions. The temporal frequency function has a low pass characteristic. Sensitivity declines above about 1 c/sec, reaching its limit at 4-8 c/sec. The spatial characteristic depends greatly on the amplitude of disparity modulation, changing from band pass at low amplitude to low pass at high amplitude. The maximum resolvable spatial frequency is 4-6 c/deg, but declines sharply for relatively high amplitudes. The interaction between amplitude and spatial frequency cannot be explained by fixed high or low limits on detectable disparity gradients.

Depth Perception↗

PubMed1996

Intrinsic uncertainty and integration efficiency in bisection acuity.

A spatial perturbation paradigm was used to determine equivalent intrinsic uncertainty and spatial integration efficiency in bisection. Specifically, three-line bisection thresholds were measured in the fovea of four normal observers with stimulus lines comprised of discrete dark dots distributed randomly around the mean line position according to a Gaussian function. The standard deviation of the Gaussian distribution (sigma e), the number (N), and the strength (C) of the dots as well as line separation were varied. Bisection thresholds were modeled by an ideal integrator, from which the magnitude of equivalent internal uncertainty (sigma i), the equivalent effective number of dots (k), and equivalent integration efficiency (k/N) were quantified. At the 2 min arc separation, sigma i decreases (down to a few sec arc) as N and/or C increases. The effects of both N and C can be accounted for by the stimulus visibility (V, in multiples of detection threshold). At the 16 min arc separation, sigma i is independent of N, C, or V, and is about 1 min arc. The two different forms of sigma i indicate that bisection judgments are limited by at least two separate sources of limiting noise, consistent with the hypothesis of two separate mechanisms (i.e. spatial filters and local signs). A visibility dependent sigma i at the 2 min arc separation can be explained on the basis of contrast sensitive spatial filter mechanisms. A fixed sigma i at the 16 min arc separation indicates a genuine positional uncertainty, consistent with local-sign mechanisms. Interestingly, equivalent integration efficiency (k/N) is very similar at the two line separations. k/N is critically dependent on, and proportional to C, indicating a common limitation in a detection mechanism.

Depth Perception↗