home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!uknet!edcastle!airdg
- From: airdg@castle.ed.ac.uk (Richard Grey)
- Newsgroups: comp.graphics.visualization
- Subject: Visual Force Detection in Landscape
- Message-ID: <28047@castle.ed.ac.uk>
- Date: 12 Nov 92 13:05:33 GMT
- Organization: Edinburgh University
- Lines: 73
-
- I hope this is the right place to post, and that somebody out there may
- be able to help me, (no matter what small a way !).
-
- The following is an introduction to a project I am undertaking for the
- Forestry Commission here in Scotland :
-
- Abstract
- --------
- The production of an automatic system
- capable of identifying the visual force
- aspects of landscape using contour data.
-
- Description
- -----------
- Visual force is a principle which is embodied in art,
- graphic design, and architecture. The eye and mind respond to visual
- force in a predictable and dynamic way. The visual forces in landform
- draw the eye down convex slopes and up concave ones, the strength of
- the force depending on the scale and irregularity of the landform.
-
- Steps
- -----
- The starting point for this project is with contour data for a
- specific region, ideally an array of National Survey coordinate /
- altitude points. From the input, we must then be able to identify the
- convex and concave features (ie. be able to label each pixel as being
- one of : valleys, ridges, hollows, peaks and saddles etc.) and then
- grouping these to form connected patches and remove trivial (single
- pixel) patches. I hope to be able to use something like the Canny
- non-maximal suppression (with Du Li's edge tracking) process for then
- finding and connection the peaks and troughs of the ridges and valleys.
-
- From all this, hopefully then we can estimate the scale and irregularity
- of the landform features and so calculate the visual saliency.
-
- ----------
-
- My first main problem is the classification of each
- pixel into either ridges, valleys etc. At the moment,
- I am looking into Besl & Jain's method of using the
- Mean and Gaussian curvatures (H & K values) to classify
- non-planar patches, and hence determine whether they
- are concave or convex, where the H & K values are found
- using a biquadratic fit.
-
- From this information, I need to group pixels into
- features so I can proceed with the Canny suppression and
- track edges.
-
- Is there any better way I might approach the two above ?
- Does anybody have any further enlightening information
- on my best approach (references seem to be virtually
- non-existent), or, better still, algorithms to do the
- above (I have some Canny stuff) ?
-
- I would much appreciate any email replies that may be of use, and, if
- there is enough interest / doubts/ queries, I will post some sort of
- summary.
-
- Cheers.
-
- Richard.
- =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
-
- airdg@castle.ed.ac.uk
- _______ __ ______ __ __ _______ _______ _______
- / ___ / / / / ____/ / / / / / ___ / / ___ / / ___ /
- / /__/_/ / / / / / /--/ / / /__/ / / /__/_/ / / / /
- / /\ \ / / / /___ / /--/ / / /--/ / / /\ \ / /__/ /
- /_/ \_\ /_/ /_____/ /_/ /_/ /_/ /_/ /_/ \_\ /______/
-
- { Edinburgh University Dept. of Artificial Intelligence }
- =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
-