Edge Detector Comparison

Heath, M., Sarkar, S., Sanocki, T., and Bowyer, K


Initial Work

The initial phase of this work was presented at CVPR '96. A copy of the paper submitted for the proceedings is available.


Comparison of edge detectors: a methodology and initial study,
Heath, M., Sarkar, S., Sanocki, T., and Bowyer, K.
Computer Vision and Pattern Recognition '96, San Francisco, June 1996.


Most Recent Work

In a newer phase of work on the comparison of edge detection algorithms, five edge detectors were evaluated. A masters thesis detailing the desription of this comparison is available. This work was published in Patten Analysis and Machine Intelligence. The reference is:

   M. Heath, S. Sarkar, T. Sanocki, and K.W. Bowyer, "A Robust Visual Method for
   Assessing the Relative Performance of Edge-Detection Algorithms" IEEE
   Transactions on Pattern Analysis and Machine Intelligence, Vol. 19, No. 12,
   December 1997, pp. 1338-1359.

The method for evaluating the performance of the edge detectors used 28 images. These images, and the edges detected in them by five edge detection algorithms are being provided for others to use. They can be downloaded by anonymous ftp from

ftp://figment.csee.usf.edu/pub/Edge_Comparison/images/results

or can be viewed and saved using a web browser.

There are 12 edge maps for each image generated by each edge detector. Each of the edge images was generated with different input parameter values. All 12 edge images can be viewed by clicking on any of the images below. Alternativly, the best edge image for each algorithm can be viewed. The best edge images were determined in two different ways; by finding the best parameters to use for the set of images (fixed parameters) and by adapting the parameters to each individual image (adapted parameters). Tables of the files best_fixed and best_adaptive reference these images for easy access to them.



EDGE DETECTOR CODE


The source code for our implementation of the Canny edge detector is being made available (see below). Recently (on 9/24/99) we made the source code for our implementation of the Rothwell edge detector available. It is listed below. We are not redistributing implementations of the other edge detectors (sorry).

The source code for the implementation of the Canny edge detector compared in this study is available.
The source code for the implementation of the Rothwell edge detector compared in this study is available.



IMAGES


(Helpful hint: If your web browser allows you to run more than one copy of it at the same time, as Netscape does, then you can use that feature to compare edge images.)

Man-made, non-textured:
140 46 147 205 218
golfcart pitcher stapler mailbox pillow

Man-made, textured:
221 208 207 214 222
brush shoppingcart tire grater picnic basket

Natural, non-textured:
212 217 219 223 225
orange banana egg elephant pond

Natural, textured:
216 201 220 203 224
pinecone feather beehive turtle tiger

Images from Experiment 1:
36 48 106 115
briefcase trashcan videocamera coffeemaker
118 124 141 149
flower airplane trafficcone stairs