Invariant recognition of natural objects in the presence of shadows.

Shadows are frequently present when we recognize natural objects, but it is unclear whether they help or hinder recognition. Shadows could improve recognition by providing information about illumination and 3-D surface shape, or impair recognition by introducing spurious contours that are confused with object boundaries. In three experiments, we explored the effect of shadows on recognition of natural objects. The stimuli were digitized photographs of fruits and vegetables displayed with or without shadows. In experiment 1, we evaluated the effects of shadows, color, and image resolution on naming latency and accuracy. Performance was not affected by the presence of shadows, even for gray-scale, blurry images, where shadows are difficult to identify. In experiment 2, we explored recognition of two-tone images of the same objects. In these images, shadow edges are difficult to distinguish from object and surface edges because all edges are defined by a luminance boundary. Shadows impaired performance, but only in the early trials. In experiment 3, we examined whether shadows have a stronger impact when exposure time is limited, allowing little time for processing shadows; no effect of shadows was found. These studies show that recognition of natural objects is highly invariant to the complex luminance patterns caused by shadows.

Human efficiency for recognizing and detecting low-pass filtered objects.

Recently, Tjan, Braje, Legge and Kersten [(1995) Vision Research, 35, 3053-3069] found that human efficiency for object recognition was less than 10%, indicating that humans fail to use much of the information available to an ideal observer. We examine two explanations for these low efficiencies: (1) humans are inefficient in using high spatial-frequency information; and (2) humans are inefficient in detecting image samples. We tested the first possibility by measuring human efficiency for recognizing low-pass filtered objects, rendered as line drawings and silhouettes, in luminance noise. Efficiency did not improve when high frequencies were removed, and the first explanation was rejected. We tested the second explanation by comparing efficiencies for object detection and recognition. Recognition efficiency was higher than detection efficiency for silhouettes but not line drawings, showing that detection efficiency does not place a ceiling on recognition efficiency. The results indicate that human vision is designed to extract image features, such as contours, that enhance recognition. A computer simulation suggests that this can occur if the observer views the world through a band-pass spatial-frequency channel.

Human efficiency for recognizing 3-D objects in luminance noise.

The purpose of this study was to establish how efficiently humans use visual information to recognize simple 3-D objects. The stimuli were computer-rendered images of four simple 3-D objects--wedge, cone, cylinder, and pyramid--each rendered from 8 randomly chosen viewing positions as shaded objects, line drawings, or silhouettes. The objects were presented in static, 2-D Gaussian luminance noise. The observer's task was to indicate which of the four objects had been presented. We obtained human contrast thresholds for recognition, and compared these to an ideal observer's thresholds to obtain efficiencies. In two auxiliary experiments, we measured efficiencies for object detection and letter recognition. Our results showed that human object-recognition efficiency is low (3-8%) when compared to efficiencies reported for some other visual-information processing tasks. The low efficiency means that human recognition performance is limited primarily by factors intrinsic to the observer rather than the information content of the stimuli. We found three factors that play a large role in accounting for low object-recognition efficiency: stimulus size, spatial uncertainty, and detection efficiency. Four other factors play a smaller role in limiting object-recognition efficiency: observers' internal noise, stimulus rendering condition, stimulus familiarity, and categorization across views.