Parameters of Recognition Algorithms for the Background Subtraction of Color Medical Images

Abstract

Mycobacterium tuberculosis infection remains a major public health issue of global morbidity and mortality. One of the widely used methods for the finding of mycobacterium tuberculosis is the Ziehl-Neelsen method of microscopy. In this paper, a method for removing noise without producing image distortion for Ziehl-Neelsen stained images of sputum smear samples obtained using a light microscope is presented. The proposed approach is based on the convolution of the original image with the Laplacian of a Gaussian filter enhanced by high-frequency filtering. Used Laplacian of Gaussian filter was discretized as a 9x9 convolution kernel. If the original image is filtered with a simple Laplacian of Gaussian, the resulting output is rather noisy. Combining this result of filtration with the enhanced by high-frequency filtering will reduce the noise and will keep of mycobacterium tuberculosis for further analysis by automated medical diagnostic systems. To deal with the automatic determination of filtering quality the normalized color difference was proposed. The developed method of background subtraction can be used for images of microscopy and dermatology.