生物谷報(bào)道：我們能夠很容易分辨出一件東西是用白蠟做成的還是用石膏做成的,，也能夠很容易分辨出一個(gè)木質(zhì)表面是否進(jìn)行了拋光處理，只要觀察表面是亮還是暗,、是否光滑就可以了,。這便涉及到一個(gè)先有雞還是先有蛋的問(wèn)題。為了推斷一個(gè)三維表面的反光質(zhì)量,，我們需要關(guān)于形狀和照明情況的詳細(xì)信息,，但要推斷其中的每個(gè)因素，又需要關(guān)于其他因素的知識(shí),。Motoyoshi等人發(fā)表在5月10日《自然》雜志的最新研究解開(kāi)了這個(gè)謎,。在此過(guò)程中，他們給出了一些對(duì)機(jī)器人視覺(jué)系統(tǒng)的開(kāi)發(fā)人員和創(chuàng)造虛擬現(xiàn)實(shí)的計(jì)算機(jī)動(dòng)畫(huà)制作人員可能會(huì)很有用的線索,。關(guān)鍵是一個(gè)簡(jiǎn)單的圖像統(tǒng)計(jì)原理：我們對(duì)光亮程度的感覺(jué)是由一個(gè)圖像中亮度值分布的正偏（positive skew）量決定的,。
FIGURE 2. Perceived lightness and glossiness may be based on the skewness of the luminance histograms.
a, The stucco-like surface on the right looks darker and glossier than one on the left even though the mean luminance of both images is equal. We note that the luminance histogram is negatively skewed for the left image (skewness = -1.34), and positively skewed (skewness = 2.40) for the right image. b, In the left panel, as the diffuse reflectance increases, the lightness ratings given by human observers increase (red circles, y-axis scale on the left side) and the skewness of the image histogram decreases (black circles, y-axis scale on the right side). The data were obtained with medium-glossy surfaces. In the right panel, we see that as the specular reflectance increases, both the rated glossiness (red circles) and the histogram skewness (black circles) increase. The data were obtained with dark-grey surfaces. c, The rated lightness (left panel) and glossiness (right panel) of the 63 surface images of varying diffuse and specular reflectance under different illumination conditions. The human ratings correlate well with the skewness of the image histogram. Error bars represent 1 s.e.m across six observers. d, The effect of histogram manipulation on perceived lightness and glossiness. For images of stucco, crumpled paper, and fabric, the luminance histogram was matched to a positively or negatively skewed beta distribution. The rated lightness (left panel) and glossiness (right panel) varied according to the final skewness of the manipulated image (mean = 8.2 cd m-2, s.d./mean = 0.1, See Supplementary Data B for data obtained with other parameters). Error bars represent 1 s.e.m. across seven observers.
原文出處：
Nature Volume 447 Number 7141
Image statistics and the perception of surface qualities p206
Isamu Motoyoshi, Shin'ya Nishida, Lavanya Sharan & Edward H. Adelson
doi:10.1038/nature05724
First paragraph | Full Text | PDF (1,022K) | Supplementary information
See also: Editor's summary | News and Views by Landy
作者簡(jiǎn)介：
Edward H. Adelson
Professor of Vision Science
Dept. of Brain and Cognitive Sciences
Massachusetts Institute of Technology
Web: http://web.mit.edu/persci/people/adelson
Secondary Office:
Computer Science and Artificial Intelligence Laboratory (CSAIL)
My lab is part of the Perceptual Science Group in the Department of Brain and Cognitive Sciences at MIT, and focuses on topics in human and machine vision, including mid-level vision, material perception, image statistics, and motion analysis.
· Research Interests
· Publications
· Illusions and Demos
· Curriculum Vitae
· People
· Patents
