Measuring tree-ring parameters using the generalised fisher ratio

A new technique is proposed to locate, count and measure the modes of a well-separated multimodal density distribution. It is based on a generalisation of the Fisher ratio for Gaussian mixture models. Under the assumption that the optimal mixture model that matches the right number of modes in a well-separated multimodal distribution is the one that maximises the Fisher ratio, we can find the optimal mixture model using a merging technique. We apply this new criterion to count, locate and measure tree-rings in dendrochronology and compare it to the most widely used method, the Hybrid Edge Detection algorithm.