Publication details

 

Improved statistical edge detection through neural networks

Basic information
Original title:Improved statistical edge detection through neural networks
Authors:Ian Williams, David Svoboda, Nicholas Bowring, Elizabeth Guest
Further information
Citation:WILLIAMS, Ian, David SVOBODA, Nicholas BOWRING a Elizabeth GUEST. Improved statistical edge detection through neural networks. In 10th Conference on Medical Image Understanding and Analysis. Manchester: BMVA, 2006. s. 56-60, 5 s. ISBN 1-901727-31-9.Export BibTeX
@inproceedings{639086,
author = {Williams, Ian and Svoboda, David and Bowring, Nicholas and Guest, Elizabeth},
address = {Manchester},
booktitle = {10th Conference on Medical Image Understanding and Analysis},
keywords = {edge detection; neural networks; statistical tests},
language = {eng},
location = {Manchester},
isbn = {1-901727-31-9},
pages = {56-60},
publisher = {BMVA},
title = {Improved statistical edge detection through neural networks},
url = {http://www2.wiau.man.ac.uk/miua2006/},
year = {2006}
}
Original language:English
Field:Use of computers, robotics and its application
WWW:link to a new windowhttp://www2.wiau.man.ac.uk/miua2006/
Type:Article in Proceedings
Keywords:edge detection; neural networks; statistical tests

The paper details a novel and successful method for multi-statistic edge detection. The detector works by analyzing the texture properties of different regions within an image, and through the use of neural networks classifying the location and direction of any edges. The detailed technique is illustrated for use both on Histological Mouse Embryo Atlas (MA) images, and also real image data. The overall accuracy of this novel technique is extensively tested using a novel grey-scale performance measure (GFOM) which allows a robustness in the results unavailable with visual inspection alone. The filter is illustrated to outperform the traditional Canny edge detector which is seen as the benchmark for edge detection. The technique presented within the paper can be applied to a variety of low level medical imaging applications and is particularly suited to images containing high levels of noise and texture where the traditional methods of edge detection prove less successful.

Related projects: