| Urinary sediment inspection is an important method of clinical diagnosis. Nowadays, it is carried out by the clinical doctors by hands and naked eyes, which is a tiered job, and impossible to get standardized. Luckily, as the development of the pattern recognition and computer vision technique, it is possible to make the inspection automatic. Generally, the automatic-inspection can be divided into three processes, the image capture and segmentation, feature extraction, and feature based image recognition. This thesis will put great emphasis on the image capture and segmentation.As for the image capture and preprocessing, the auto-focus and preprocessing algorithm are considered to obtain well-captured image, which is the basis for the reliable image segmentation. The urinary sediment image is captured by microscope, the depth-of-field of which is relatively low, consequently it is easy to lose focus and get blurred image. Thus, an auto-focus algorithm based on the high-frequency of the image is investigated to easily get the well-focused image before capturing. Besides, several preprocessing steps are conducted on the captured image before the segmentation to improve the quality of the image.As we have known, there are two types of segmentation method, that is to say, edge-based and region-based, both of which pay total attention to the foreground. In this thesis, we solved the issue in a new way. The pixels of the image can be divided into two types: foreground pixels and background pixels. The scope of the background pixel intensity can be computed using the Gaussian Model of the background. Using the model we developed a new double-threshold segmentation method, which is used in both the intensity image and gradient image. The result shows that the sediment (cell, crystal and so on) in the image can be segmented exactly and the total detection rate is 98%. The new segmenting method is speciallysuited to the segmentation of the image whose histogram has single peak.
|
PDF Full Text Request