Journal of Computational Interdisciplinary Sciences (2011) 2(3): 173-177 © 2011 Pan-American Association of Computational Interdisciplinary Sciences Printed version ISSN 1983-8409 / Online version ISSN 2177-8833 http://epacis.net doi: 10.6062/jcis.2011.02.03.0044 Poisson noise reduction in deconvolution microscopy Murillo R.P. Homem 1 , Marcelo R. Zorzan 2 and Nelson D.A. Mascarenhas 3 Manuscript received on December 21, 2011 / accepted on December 30, 2011 ABSTRACT Computational optical sectioning microscopy is a powerful tool to reconstruct three-dimensional images from optical two-dimensional sections of a biological specimen acquired by means of a fluorescence microscope. Due to limiting factors in the imaging systems, the images are degraded by both the optical system and detection process. Each of the two-dimensional section of the three-dimensional data set are blurred by contributions of light from other out-of-focus planes. Besides, they are also corrupted by noise due to quantum fluctuations of light. In this work we present a method to perform the restoration of three-dimensional data obtained by fluorescence microscopy. The algorithm consists of the use of a noise reduction procedure based on the Anscombe transformation followed by the Richardson-Lucy deconvolution algorithm. Results showed an improvement on deconvolution performance when using phantoms and real cell images. Keywords: computational data analysis and simulation in general sciences, computational optical sectioning microscopy, deconvo- lution microscopy, poisson noise, Anscombe transformation. 1 INTRODUCTION The visualization of the proper three-dimensional (3D) image of a biological specimen is important in many specific problems because the cell structure and its function can be strongly corre- lated. This aim can be achieved with the use of the computational optical sectioning microscopy (COSM) technique. In COSM, a 3D image is formed by stacking a series of two-dimensional (2D) images that are acquired by using light microscopy. With the use of a widefield or confocal microscope it is often used in fluorescence microscopy where the data represents the fluores- cence concentration [1]. However, the 3D images acquired using COSM are degraded by the microscope optics. Indeed, each slice of the 3D image is blurred, i.e., it has contributions of light from other out-of-focus planes [1, 2]. As a result, the light originated from a point in the focal plane is not exactly imaged as a point. Also, the Fourier op- tics theory demonstrates that there is a cut-off spatial frequency which is directly determined by the shape and size of the limiting pupil in the optical system. In this sense, some image content are lost during image acquisition [3]. Besides the blurring due to the microscope optics, there are several sources of noise that decrease the quality of the images [1, 2]. As a result of the many problems that arise when imaging living cells (for instance, to avoid photobleaching), the images are often recorded under low-level light. In these situations, the images are quantum limited and they are corrupted by Poisson noise. Other noise sources come from the charged-couple device (CCD) camera systems that are usually used to record the images. For instance, thermal noise due to electronic devices. Correspondence to: Murillo R.P. Homem – E-mail: murillo@ufscar.br 1 Departamento de Computac ¸˜ ao de Sorocaba, Universidade Federal de S˜ ao Carlos, campus de Sorocaba, Sorocaba, SP, Brazil. 2 Universidade Federal de Vic ¸osa, campus de Rio Parana´ ıba, Rio Parana´ ıba, MG, Brazil. 3 Departamento de Computac ¸˜ ao, Universidade Federal de S˜ ao Carlos, S˜ ao Carlos, SP, Brazil.