High Dynamic Range Image Processing of Confocal Laser Scanning Microscopy Data

Abstract

All imaging sensors suffer from having a limited dynamic range, that is, the range of light intensities which can be accurately measured simultaneously, thus potentially limiting the amount of detail which can be captured in a single image. A method for generating high dynamic range (HDR) images from several regular low dynamic range (LDR) images captured using a photomultiplier tube imaging sensor for a confocal laser scanning microscope has been developed. This method uses data from all the available images in order to generate images which simulate an imaging sensor having unlimited dynamic range and simultaneously reduces the noise in the pixel intensity observations. The algorithm is based on solving a maximum likelihood estimation problem for censored data which yields estimates of the true, non-censored intensity values. The proposed method is demonstrated to generate HDR images containing details which were difficult or impossible to observe in the LDR images for several different kinds of samples. Furthermore, the method is validated by introducing artificial censoring in the original data and comparing the estimated intensities for different artificial thresholds to the observed intensities. The proposed method is shown to accurately predict the non-censored pixel values even for very low artificial thresholds (effectively simulating the behaviour of a sensor with a reduced dynamic range) which indicates that the method also accurately predicts the truly censored pixel intensities.