Home Home
Introduction Introduction
Goals Goals
Our model Our model
2D results 2D results
3D results 3D results
Conclusion Conclusion
References References
Word-book Word-book
francais english
Introduction Introduction Goals Goals Our model Our model

Goals of this work - Proposed Methodology

The Color Infrared aerial photographs, once scanned at a resolution of 50cm/pixel, represent precious data for the image processing community. The aim is to design some algorithms which would give automatically or semi-automatically some resource parameters by processing the image, and extracting its components. Then, statistics such as the number of trees, their size or the density of the stands could be obtained and give some useful information at the scale of the tree which are not conceivable without any help of image processing (it would take too much time for an operator to get these statistics, and also be very costly).


We propose to use marked point processes publi to extract the tree crowns from forestry images. They enable to model an unknown number of geometrical objects in a scene within a stochastic framework. They are interesting because they can embed both some prior information about the interaction between the features to be extracted (all the knowledge we have on the forest), and some data information to fit them in the image. The extraction result gives us the positions of the trees and their size.


We designed a 2D model and a 3D model which are adapted to respectively dense areas and sparse vegetation. Indeed, the two information which help us to extract the tree crowns are the high reflectance of the trees in the near infrared and their shadow, which can be located either all around the crown in dense areas, or just in the direction of the sun (drop shadow) in sparse vegetation for instance. In the latter case, a 3D model is used because we can get the height of the tree from the size of the shadow.


Guillaume Perrin - icone mail XHTML valide XHTML valide