Study On Methods Of Leukocyte Image Segmentation Based Fingerprint

In recent years, automatic identification for the blood leukocytes of medical image applications has become an important topic. Automatic detection and identification for the leukocytes has a high reference value to quickly and accurately determine whether have the disease or the disease type and extent of serious in clinical practice. At the same time, automatic identification for white blood cells also has a good effect in improve efficiency and significantly reducing manual operation and reduce the objective factors of human error. Cell image analysis is the first and crucial step for the cell automatically identify, the quality of segmentation will directly affect the accuracy and classification for the parameters extraction and the final recognition rate for the classification system.In this paper, with the characteristic of the image of human peripheral blood leukocytes, it has an in-depth research for the automatic segmentation of the color leukocytes. First, read the image of the original histogram as the fist step, but for the interference of the "spike" pulse, result many peaks and valleys areas in the original histogram. And then carrying on the multi-criterion spatial filtering to the primitive histogram, obtains the image criterion free-space diagram. From the first and second zero crossing location information in scale space map can be obtained fingerprint, analysis the fingerprint feature to determine the optimal value of scale space. It can achieve a good smoothing effect, using Gaussian filter for the optimal scale values. After smoothing the image not only smoothed out the noise, while maintaining the details of the part. After the smoothing for the cell image, searching for an appropriate threshold value to segment the image, can obtain the cell nucleus image. The extraction of cytoplasm, binary the image in the first, and then do the distance transformation with it, to transform the distance transformation into the pixel information, and then using the adaptive watershed algorithm to distinction different element spots, find the adhesion boundary points, thus separated the adhesion cell, experiments show that the algorithm can successfully separate the cell adhesion.Using the segmentation algorithm proposed in this paper segmentation of the five types of leukocyte image processing, experimental results show that the segmentation algorithm using in this paper have a good effect in accuracy, speed and cell division to maintain the edge, and have a 91.95% recognition rate in the segmentation of the total five cells. Finally, the nucleus were segmented using the mathematical morphology, so that it can receive a more complete and more accurate image segmentation, and also it is conducive to the analysis of medical and clinical disease more accurate diagnosis.