Cite as:
Mamassian P, Hérault J, 1999, "Perception of the illumination of a natural scene by man and machine" Perception 28 ECVP Abstract Supplement

Perception of the illumination of a natural scene by man and machine

P Mamassian, J Hérault

Changes in illumination conditions can have a drastic effect on the appearance of a scene. These effects are both global across the whole scene and local in the neighbourhood of shadow boundaries. We focus here on the global problem of inferring the illumination direction from both psychological and computational standpoints. In a 2AFC experiment, observers had to decide whether a natural scene was illuminated from the left or the right. Original grey-scale images were intermixed with images combined with a luminance ramp from left to right. Observers were faster and more accurate when the luminance ramp was consistent with the light direction in the original image, and the converse when it was inconsistent. These results indicate that the low spatial-frequency content of an image is critical for the perception of the illumination direction. We then looked at a simple model that could extract the illumination direction from natural images. The model exploits the ON and OFF signals at the level of the retinal ganglion cells for very low spatial frequencies; the 2-D cross-correlation between these two signals provides an estimate of the illumination direction. The model's behaviour is compared with human performance, and its biological plausibility is discussed.

These web-based abstracts are provided for ease of seaching and access, but certain aspects (such as as mathematics) may not appear in their optimum form. For the final published version of this abstract, please see
ECVP 1999 Abstract Supplement (complete) size: 1478 Kb