Abstract:
Tuberculosis is one of deadly disease in the world especially in developing countries. Sputum smear microscopy is the main diagnostic tool in developing countries and high TB burden countries for the detection of TB .previous studies show that the manual screening for sputum smear microscopic images lead to misdiagnosis and false result, Image processing techniques are applied in this research to enhance, segment and classify the sputum smear images for computerized process of TB bacilli identification. Image processing algorithms which used for sputum smear image include a series of enhancement techniques, segmentation methods and morphological operation. As the non-bacillus objects in sputum smear image can bias the detection, it should be suppressed from the smear image. This research employs color image segmentation technique for the segmenting the TB bacillus objects from the background.
TB bacillus objects are segmented from the background in two stages based on color space conversion and k-means clustering ,to identify the TB bacilli, the proposed method uses eleven shape feature descriptors, which are compactness, eccentricity ,area, perimeter and Hu moments (𝑀1𝑡𝑜𝑀7),and makes the judgment using a support vector machine .experimental result confirmed the superior performance of the proposed method .