Research Of Peripheral Blood Cell Image Segmentation Algorithms

A medical operator needs to carefully observe peripheral blood and bone marrow samples for diagnosis of diseases under a microscope,including the number of blood cells and the cell form.This process is very time-consuming and is affected by the subjective factors of the operator.Blood cell parameters of statistics and analysis are important in clinical test,therefore,exploring a new detection method for peripheral blood cell is important by making use of image processing techniques.In this dissertation,according to the characteristics of the peripheral blood cell image,the key techniques using the method of digital image processing and analysis,which contain peripheral blood cell contour extraction,concave point detection and image segmentation,are discussed in details.Firstly,the improved algorithm based on an active contour model is presented to extract the contours of peripheral blood cells.Secondly,considering the difficulty of detection for touching-cells,this paper puts the characteristics in touching-cells into the normalized cut standards for building reasonable weight matrix,which is applied into the image.The concrete research contents are as follows:1.Image preprocessing.The acquisition of peripheral blood cell images is a complicated process,there are many affecting factors including the noise of image and uneven illumination in the process,which will increase the difficulty of subsequent image processing,so the preprocessing is necessary.Most of acquired images are color level while the previous algorithms are mainly suitable for gray level images,therefore,the color transformation is implemented.Then,according to the categories of noises in images,the different filter algorithms are chosen.2.Cells' contours extraction in peripheral blood cells images.This paper puts forward the improving energy function of the traditional Snake model.The external field based on the control points is applied,and the force function is joined to enhance the strength.The expansion of the Laplace operator is applied to the image of gradient amplitude.The new Snake algorithm can restrain interference well and extract contours with peripheral blood cells of the image.The complete contour provides the possibility for subsequent touching-cells division.3.Main concave points extraction.The paper proposes the searching rules based on concave points in touching-cells,and the pixels on the edge of the touching-cells can be divided into concave points and non-concave points.The concave points determination rule is to classify the concave point based on the proportion of line segment.The search for clarity is practical,and can accurately extract the main concave point in touching-cells,which is the premise to the subsequent touching-cells division.4.Touching-cells splitting.In this paper,the improved normalized cut algorithm is proposed based on the main concave points and the discriminant segment,aiming at the relevance between concave points and pixels.And the gradient value takes the place of the gray value in a traditional build matrix.Its good use of the concave point information plays a role in image segmentation,improving the segmentation accuracy and avoiding the segmentation error.