Automatic Color Blood Cells Image Segmentation Based On ELM Algorithm

In computer aided diagnosis, people often used blood cell of morphology and the number of cell in the blood for histopathologic diagnosis. In the past, the diagnosis is done by manual, In the process, people tend to visual fatigue, at the same time it impact the accuracy of diagnosis that due to the subjective factors of diagnostician such as emotion, attitude, fatigue. Nowadays with the development of computer technology, we use the computer to deal with blood cell images, which can be very accurate and reasonable to extract the image information of the blood cells such as cell morphology and cell number. It is more accurate to diagnosis blood disease when we had this information of cells. So it can reduce the visual fatigue of diagnostician and can improve the diagnostic accuracy for use of computer-aided diagnosis. People want to extract cell information and count cells. The first thing is to segment cells. So accurate cell segmentation is the premise to realize blood disease diagnosis.This paper proposes a method to extract nuclei based on maximum entropy threshold, positioning to the area of interest, according to the interested region to regional expansion. Then, we put the image segmentation as two kinds of classification problem, the corresponding sample selection strategy is used to sample in interested area and background area, using classification algorithm of extreme learning machine (ELM) to training classification model. Finally in order to achieve the purpose of automatic cell image segmentation, it can use classification model to classify cell image pixels. The paper main research content is as follows:(1)According to the prior knowledge that the nuclei of blood cell image is darker, through the comparison experiment, using the threshold algorithm based maximum entropy to locate interested area-the nucleus, the advantage of this method is fast, accurate positioning. We are looking for a kind of method to collect the training sample of the pixel in high gradient automatic, collecting the pixels that contains large amount of information of as the training sample set, which can reduce the running time, and improve the algorithm performance.(2)The problem of cell image segmentation is converted to two classification problem for image pixels. Because the ELM classification algorithm can study with small sample and have good generalization performance of the machine learning algorithms, so this article use the fast classification algorithm based ELM to classify the pixels. At the same time, according to the characteristic of ELM classification is not stable, multiple classification model of ELM is integrated into a single total classification mode through use of Bagging integration strategy, so as to improve the stability of classification.(3)In this paper, Study a method for an image real-time sampling, at the same time automatic segment the same image. In this paper, according to the characteristics of the ELM algorithm classification speed, adopt to each real-time segmentation of cell image sampling training classification model, and then use the classification model to segment the cells image of sampled. This whole process without any artificial participation and parameter adjustment, just giving a cells image which can be segmented is automatic divided to achieve real-time change fully automated blood cell image segmentation.Through experiments on100blood cell images, the automatic segmentation method for blood cell images proposed in this paper has been proven to have the characteristics of high speed and accuracy, no artificial participation, etc.