| Abstract: |  This thesis has two contributions:
First, a new semi-automatic method for conifers identification based on images of needles (or more precisely images of coniferous branches) was proposed.
The method's inputs are (except of the image) the coordinates of two branch-points.
After finding the branch shape and the "needle area" in the branch surroundings, the texture pattern direction is described and normalised to correspond the branch direction and shape. The descriptor is then a 4-D vector (histogram) and the nearest neighbour classifier is used for identification. A dataset containing 93 images of 4 species (fir, spruce, larch, pine) was created. An experiment on this dataset shows 72.3% classification precision.
Second, a mobile application for plant identification was created, which allows users to identify plants using image recognition (from leaves and bark, as proposed by Sixta, and from needles, as described in this thesis), manual field guide or reading plant species characteristics. This application also allows to save the photographed images (with GPS coordinates) or upload them to a community website, which was also created within this thesis.
|
|---|