Automatic Calibration of the Deceived Non Local Means Filter for Improving the Segmentation of Cells in Fluorescence Based Microscopy

This paper presents an automatic approach for optimal calibration of the deceived non local means filter (DNLM), for enhancing segmentation accuracy of fluorescence based microscopy images. The DNLM is designed for image denoising and enhancement. The calibration of its parameters in real image preprocessing applications is very often time consuming, since doing a manual calibration in a sample image might not work in different image samples. We compared three different stochastic optimization approaches: simulated annealing, particle swarm optimization and genetic algorithms, and selected the best approach. The implemented solution needs from the user only to define a precision metric and a set of image samples, and the algorithm will arrive to a locally optimal set of the filter parameters, to improve segmentation accuracy, using Otsu thresholding and measured with the Dice index. The PSO approach presented the overall best performance, with an average Dice index of 0.9667 in the validation set, a two percent boost over the best manually calibrated set of parameters for the DNLM.