Feature Requests: Searching and Goal Oriented Evaluation

After talking with Dr. Doermann, I've added a few new feature requests to the ViPER project. One is for some additional api functionality that MALACH requires, a metadata query system. The other is for performance evaluation, although it will likely require more semantically meaningful ground truth as well; it is a feature request for goal-oriented evaluation support.

Link Link

- posted by David @ 6:50 PM

Analysis of PCA-based Face Recognition Algorithms

This paper uses the FERET '96 protocol to deal semi-systematically evaluate different parameters to a modular PCA-based face detection system.

Reference Link

@article{Moon1998
   author = {Hyeonjoon Moon and P. Jonathon Phillips},
   title = {Analysis of PCA-based Face Recognition Algorithms},
   book = {Empirical Evaluation Techniques in Computer Vision},
   year = {1998},
   pages = {57--71}
}

- posted by David @ 11:18 AM

Performance Evaluation of Clustering Algorithms for Scalable Image Retrieval

This paper presents an evaluation of an image retrieval system. Since it is only a performance optimization of an existing system, the evaluation compares the new system results to the old system. As such, there is only a retrieval accuracy, and not seperate precision and recall (as the size of the sets are equal). As a whole, this paper is more a presentation of the clustering technique than of an evaluation method.

Reference

@article{AbdelMottaleb1998
   author = {Mohammed Abdel-Mottaleb, Santhana Krishnamachari, and Nicholas J. Mankovich},
   title = {Performance Evaluation of Clustering Algorithms for Scalable Image Retrieval},
   book = {Empirical Evaluation Techniques in Computer Vision},
   year = {1998},
   pages = {45--56}
}

- posted by David @ 4:55 PM

A Blinded Evaluation and Comparison of Image Registration Methods

Fitzpatrick and West evaluated a set of retrospective, meaning from pure image data, medical scans of human heads against a gold standard prospective method, meaning one that required markers placed on the patient's skull. This required a human photoshopping out the markers. Like the VACE analysis so far, they gave out some sample data with ground truth attached, and a larger sample of images without ground truth for actual evaluation.

It is also worth noting that the article uses 'blinded' to mean that the client sites, who generated the retrospective (candidate) data, did not have access to the ground truth until after the evaluation. As I mentioned earlier, this is similar to the VACE evaluations. For more information about that, see Mariano &c's paper.

Reference

@article{Fitzpatrick1998
   author = {J. Michael Fitzpatrick and Jay B. West},
   title = {A Blinded Evaluation and Comparison of Image Registration Methods},
   book = {Empirical Evaluation Techniques in Computer Vision},
   year = {1998},
   pages = {12--27}
}

- posted by David @ 2:57 PM

Empirical Evaluation Techniques in Computer Vision

So, one of the books I found while doing a literature search was Kevin Bowyer and P. Jonathon Phillips's Empirical Evaluation Techniques in Computer Vision. The text, a loosely edited book from 1998, has a lot of moderately interesting papers about analysis of specific vision techniques. It does show its age, and in fact seems much older than even that; many papers cover low level operators like edge detection and image registration. I'll try to cover each article in turn. (It should be noted that I already covered Atul Chhabra and Ihsin Phillips's article about benchmarking their CAD recognition system.

Link Reference

@book{Bowyer1998
   author = {Kevin W. Bowyer and P. Johnathon Phillips},
   title = {Lecture Notes in Computer Science},
   editor = {Kevin W. Bowyer and P. Johnathon Phillips},
   year = {1998}
}

- posted by David @ 2:12 PM