![]() | Only 14 pages are availabe for public view |
Abstract Human blood cells consist of Red Blood Cells ‘RBCs’, White Blood Cells ‘WBCs’, and platelets. The red blood cells are responsible for the oxygen transportation to all cells and tissues, and transmission of carbon dioxide to the lungs. The white blood cells have several types, which are essential in the human body’s immune system for protection from viruses, germs and bacteria. The microscopic images of the blood cells (BCs) have different nature, shape, and color. Each type of BCs in the microscopic images provides information that helps in the diagnosis and treatment of various diseases, such as leukemia, malaria, and anemia. The BCs segmentation has an impact role in the diagnosis and treatment such diseases. It is also a crucial stage in image processing steps for successful further stages, including feature extraction, and classification. However, blood cells segmentation to determine the region of interested (ROI) is considered a challenging task due to the cells’ complex shape/ nature (texture, color, shape, and size), the ambiguity, uncertainty, and inconsistencies in the microscopic captured images due to the varying illumination and the existence of overlapped cells. This thesis proposed a new weighted transformation-based segmentation method to detect five types of BCs. This method depends mainly on the discrete cosine transformation technique (DCT), which deals with real values instead of complex values, reduces the computation complexity, enhances the image contrast, and reduces the illumination variance in an image by removing the low-frequency in the DCT coefficients. A threshold was applied to the proposed weighted segmentation to determine the ROI. Different evaluation metrics were calculated to assess the performance of the proposed segmentation method. Moreover, a comparative study with other well-known methods was conducted. The results proved the superiority of the proposed weighted transformation–based segmentation method, which achieved average segmentation metrics over the different five types: 99.8% accuracy, 99.9% specificity, 89.3% sensitivity, 91.8% Dice and 85.3% Jac. |