Finding an optimal match window for spruce top detection based on an optical tree model

A method for detecting Spruce tree tops in high resolution panchromatic ima ges at acute angles has earlier been reported. The method uses an optical t ree model and flight data to generate a template tree image, i.e., an image of what the "average" tree would look like if placed at a certain point in the imaged scene. A window into the template image is matched to the actua l forest image by at each pixel computing the correlation of the local inte nsity values. Local maxima in the resulting correlation image correspond to high similarity with the template image and thus correspond to positions w here there might be a tree. It is clear that the shape and size of the matc h window and its placement within the template image has an impact on the p erformance of this tree top detection algorithm. For a specific parameteris ation of the match window shape, size and placement, and a specific set of images where the real tree positions are known one can find an empirical op timum for the average performance of the algorithm over the images. Such an examination for images of an even-aged pure Spruce stand yielded surprisin g results: the optimal size was significantly smaller than expected and the optimal shape close to a circle. By using the estimated optimal parameter values rather than those used in the work earlier reported, the tree top de tection rate was increased from 85-92% to 91-98%.