→ skylikewater:bias會不會是預期 或動機之類的? 06/18 15:39
推 brendonfish:從報導看來，滿類似 Laurent Itti 的 attention model 06/25 14:03
→ brendonfish:但處理了更多的圖型特徵。 06/25 14:05
Computer Intelligence Predicts Human Visual Attention for First Time
ScienceDaily (June 17, 2010) — Scientists have just come several steps
closer to understanding change blindness -- the well studied failure of
humans to detect seemingly obvious changes to scenes around them -- with new
research that used a computer-based model to predict what types of changes
people are more likely to notice.
These findings on change blindness were presented in the Journal of Vision.
"This is one of the first applications of computer intelligence to help study
human visual intelligence, " said author Peter McOwan, professor at Queen
Mary, University of London. "The biologically inspired mathematics we have
developed and tested can have future uses in letting computer vision systems
such as robots detect interesting elements in their visual environment."
During the study, participants were asked to spot the differences between
pre-change and post-change versions of a series of pictures. Some of these
pictures had elements added, removed or color altered, with the location of
the change based on attention grabbing properties (this is the "salience"
level referred to in the article).
Unlike previous research where scientists studied change blindness by
manually manipulating such pictures and making decisions about what and where
to make a change, the computer model used in this study eliminated any human
bias. The research team at Queen Mary's School of Electronic Engineering and
Computer Science developed an algorithm that let the computer "decide" how to
change the images that study participants were asked to view.
While the experiments confirmed that change blindness can be predicted using
this model, the tests also showed that the addition or removal of an object
from the scene is detected more readily than changes in the color of the
object, a result that surprised the scientists. "We expected a color change
to be a lot easier to spot, since color plays such an important role in our
day-to-day lives and visual perception," said lead researcher Milan Verma of
Queen Mary.
The authors suggest that the computer-based approach will be useful in
designing displays of an essential nature such as road signs, emergency
services, security and surveillance to draw attention to a change or part of
the display that requires immediate attention.
"We live in a world in which we are immersed in visual information,"
explained Verma. "The result is a huge cognitive burden which may hinder our
ability to complete a given task. This study is an important step toward
understanding how visual information is processed and how we can go about
optimizing the presentation of visual displays."
Story Source:
The above story is reprinted (with editorial adaptations by ScienceDaily
staff) from materials provided by Association for Research in Vision and
Ophthalmology, via EurekAlert!, a service of AAAS.
Journal Reference:
1. M. Verma, P. W. McOwan. A semi-automated approach to balancing of
bottom-up salience for predicting change detection performance. Journal of
Vision, 2010; 10 (6): 3 DOI: 10.1167/10.6.3
--
原始網址：
http://www.sciencedaily.com/releases/2010/06/100616171720.htm
大意是利用人工智慧來預測當一個景緻進入視覺後，
視覺區會注意哪些地方。
但詳細的我看不是很懂，只大概知道這個模型是第一個用人工智慧做的，
測試材料是一堆景緻增加或減少某些東西，或者改變物體顏色。
之前的實驗都由人決定要改變哪裡，而這一個是用人工智慧去計算要改變哪裡，
不會有人類判斷的bias。但對於這一句話我不是很能理解就是了，人類的bias會造成哪種
影響？
最後他希望這可以用來改善設計路標，或緊急設施等。
--
※ 發信站: 批踢踢實業坊(ptt.cc)
◆ From: 140.112.33.135