Automated detection and identification of abnormal cells in the human body is a critical application for medical image computing. Enhancement and de-noising of images remain challenging tasks and imperative steps for image analysis algorithms. Indeed, due to its early role in the process, the results of advanced operators for feature extraction will highly depend on the quality of enhanced image produced. Depending on the presence of different noise types, particular algorithms will respond better. This paper presents a comprehensive comparison between several linear and non-linear filters applied on fluorescence microscope images for the localization and counting of specific cancer phenotypes from mouth cell samples. The objective analysis proposed is evaluating the PSNR and Delta-SNR (the SNR to SNR measure between original images and filtered ones) for blood sample sequences taken from Cancer Research Malaysia. Thirty Fluorescence microscope images with low contrast and non-uniform illumination have been tested and analysed. Non-linear algorithms seem to show improved contrast and background removal abilities compared to linear blurring and approximating filters. © Springer Science+Business Media Singapore 2017.