APGD Virtual Lab -- Remote Execution

Linear Delineation

NOTE: The "algorithm" available on this Virtual Lab page is part of the Linear Delineation system presented in Ref: M.A. Fischler, "Finding the Perceptually Obvious Path," DARPA Image Understanding Workshop, May 1997 (see Fig 1). It is designed to operate under the condition that the line structures are coherent and clearly visible (without searching) to a naive human observer. Edges are not detected by this algorithm.

While in theory, the line structures the algorithm is able to deal with can have arbitrary width, performance will generally degrade when the lines to be delineated exceed 5-6 pixels in width. The ideal situation is where the the original line resolution is greater than the above width range and the image has been Gaussian-smoothed (using a filter with a width approximately equal to the line width); it can then be decimated ("zoomed") to obtain the image to be processed.

The Road Delineation application invokes a semantic model that looks for simple roads in a rural context. It is not currently designed to deal with the semantics of city streets, freeways, multi-level interchanges, etc. When such "road-like" or road-related objects are present in the image, the algorithm will often do a credible job in delineating them, but its performance with respect to such objects can be inconsistent.

Images submitted to be processed can be in any standard format, but at present, it is assumed they will be limited in size to a maximum of 500,000 8-bit pixels.

Select an image:
Your image. Enter URL below:

Sample image. Select name below:

Some acceptable image formats are: GIF and JFIF (a.k.a JPEG). (Actually, any 8-bit image that can be converted by the program "anytopnm" and has been assigned the mime-type "image" by the HTTP server should work. )
What type of image is it?
Panchromatic SAR
Zoom image?
No
Smooth image?
No Yes
Enter your e-mail address:

A notification that the processing is complete will be sent to this address.
Which results would you like?
1. Mask
  Detection sensitivity: Normal High
  Path characteristics: Bright lines Dark lines
2. Generic Paths
  Path visibility: Normal Very Good
  Minimum Segment Length Retained (path extraction only):
  Short Normal Long
3. Road Segments and Connectivity
When would you like this to run?

Now.
Tonight.

To initiate processing, press this button:

Aaron Heller <heller@ai.sri.com>

Created: Mon Jul 7 1997
Version: $Id: ld-exp.html,v 1.1 1997/10/10 18:29:31 heller Exp $