Comments
http://stuartjchi.blogspot.com/2011/03/paper-reading-18.html
http://shennessy11.blogspot.com/2011/03/p.html
Reference Information
http://stuartjchi.blogspot.com/2011/03/paper-reading-18.html
http://shennessy11.blogspot.com/2011/03/p.html
Reference Information
Title: Evaluating Automatic Warning Cues for Visual Search in Vascular Images
Author: Boris W. van Schooten, Betsy M.A.G. van Dijk, Anton Nijholt, Johan H.C. Reiber
Publisher: IUI '10, February 7-10, 2010 Hong Kong
Summary
The researchers in this paper focused on performing computer-aided visual search. Traditional visual search tasks include finding weapons in x-rayed baggage, targets from a moving vehicle, aerial photographs, cancer areas in mammograms, polyps in colonoscopy, or low credibility areas in automatic medical image segmentation. In many cases, automatic warning systems are used to highlight potential targets. Although, these systems can be imperfect and result in false positives/alarms. Not only that, human error also results due to over or under-reliance on such systems.
To counter this, researchers decided to make a system that alerts the human analyzing the picture for possible areas of interest, letting them decide whether they need to proceed further. They found that the user generally preferred paranoia alerts which typically display false positive instead of a more conservative system. Users performed significantly faster with paranoid highlighting than with no highlighting, and they make significantly less errors.
Discussion
This paper was fairly technical, similar to the last one. It was easy to grasp what they were going for. I thought that the problem they wanted to look at was fairly straightforward, almost obvious. I'm not sure why they thought people might be more comfortable with false negatives than positives. In areas such as cancer/weapon detection, it seems as though there's almost no extent to "too much testing".
The researchers in this paper focused on performing computer-aided visual search. Traditional visual search tasks include finding weapons in x-rayed baggage, targets from a moving vehicle, aerial photographs, cancer areas in mammograms, polyps in colonoscopy, or low credibility areas in automatic medical image segmentation. In many cases, automatic warning systems are used to highlight potential targets. Although, these systems can be imperfect and result in false positives/alarms. Not only that, human error also results due to over or under-reliance on such systems.
To counter this, researchers decided to make a system that alerts the human analyzing the picture for possible areas of interest, letting them decide whether they need to proceed further. They found that the user generally preferred paranoia alerts which typically display false positive instead of a more conservative system. Users performed significantly faster with paranoid highlighting than with no highlighting, and they make significantly less errors.
Discussion
This paper was fairly technical, similar to the last one. It was easy to grasp what they were going for. I thought that the problem they wanted to look at was fairly straightforward, almost obvious. I'm not sure why they thought people might be more comfortable with false negatives than positives. In areas such as cancer/weapon detection, it seems as though there's almost no extent to "too much testing".
No comments:
Post a Comment