|
COUNTING TREE CANOPIES FROM LOW-LEVEL COLOUR AERIAL IMAGES |
|||
|
Not an aerial image but this is the effect - try counting the tree canopies |
Are you interested in developing novel applications of computers, working closely with medical staff on medical applications of computers, computer vision research. Do you thrive on a challenge ? This project will involve all of this and provide an opportunity to participate at the forefront of exciting research working with staff in the School of Electronic and Electrical Engineering and in Computer Science at the University of Massachusetts. |
||
|
David Pycock |
Tel.: |
+(0)121 414 4330 |
|
|
|
Email: |
D.Pycock@bham.ac.uk |
|
|
|
WWW: |
http://www.eee.bham.ac.uk/pycockd |
|
|
Background |
|
|
|
Under the international strategy for environmental monitoring there is a need to be able to count trees in the rain forests from low-level aerial images. In these images one canopy merges with the next and there is only a hint of the boundary resent. It is therefore difficult to delineate and count each tree canopy. |
||
|
Objectives |
|
|
|
|
|
To develop colour mathematical morphology algorithms. |
|
|
|
To count tree canopies in dense forests. |
|
Previous Research |
|
|
|||
|
Mathematical morphology describes set operations on binary and grey-level images that can be used to estimate the boundary, centre, skeleton and size of objects. In the case of grey-level operators this can be done without a need to detect object outlines. These operations are, in effect, a combined topological and amplitude filter. Meyer describes a colour watershed algorithm for image segmentation in which distances between points in (r, g, b) colour space are used as an analogue to amplitude in a grey-level image. Just as it is a limited view of grey-level morphology to process a series of amplitude profiles and reassemble them it is also restrictive to take the distance (in r, g, b space) between points as the measure of colour distance. Whilst grey-level operators and trainable methods for defining the neighbourhood structure element are described no method exists for performing these operations on colour images. |
|||||
|
Proposal |
|
|
|
|
|
|
In this project alternative definitions for colour morphology will be investigated, starting with the basic operations of erosion and dilation. From these operations we will define colour opening and closing. To define the computation of the watershed (a standard higher order grey-level morphology operator) in colour space new concepts based on the definition of axes in colour space will be needed. In grey-level and binary morphology erosion (dilation) is defined as a neighbourhood minimum (maximum) operator. We propose to define colour erosion (and dilation) as a set operation with respect to a 3-D surface patch. |
|||||
|
Robust estimators (such as the median) of the local surface and the distance to that surface will be considered. It is also possible that using a colour model based on the human visual system, such as L*a*b* might provide results that are easier to understand than other models. We propose to investigate trainable methods of defining neighbourhood operators. This will be more than an application of previous work (on grey-level images) because of the greater dimensionality of the data set and the different nature of the distance metric. To test the algorithms generated synthetic and natural images will be used. Synthetic images will allow the basic behaviour and the response to noise of the algorithms to be determined. Natural images well help to establish the practical value, or otherwise of the operators. A bench mark for the algorithms developed will be provided by an interactive classifier developed at the University of Massachusetts. |
|
Facilities and Benefits |
|
|
||||
|
|
|
|
|
|
||
|
Contact Information Mrs M Winkles |
|
|
||||
