Research On Segmentation And Recognition Of Microscopic Leucocytes Image

In recent decades, with the development of computer technology and image processingmethod, theories and researches on pattern recognition and artificial intelligence made theimage analysis system gradually become one of the automatic quantitative analysis inmedical field and test methods.Cell image recognition technology is a major problem in the biomedical field.Segmentation and recognition of the peripheral blood leukocytes is an important step. Thenumber and percentage of various types of white blood cells in disease and under normalcircumstances are different; doctor judge this to diagnose, the type of disease and severity ofthe underlying data for medical diagnosis of blood disease research of great value for clinicalexamination. Digital image processing techniques can help doctors in their analysis anddiagnosis.The blood cell analyzer is not precise enough for peripheral blood smears and otherdefects exist in the morphological examination by artificial visual completion. Abnormalleucocytes play an important role in detection in hematology diagnostic process andaccelerate the diagnosis of leukemia disease. Therefore, appropriate treatment of earlydisease detection is necessary. Digital image processing technique can help doctors enhancethe important features of visualization, analysis and diagnosis of leukocyte. Cellularautomata to identify the classification technique introduced into the blood clinicalexamination, can reduce the physician labor intensity and error caused by human, andimprove the accuracy and efficiency of the detection results.To accomplish the goal of leucocytes image recognition and classification need to gothrough a series of steps. For24-bit true color microscopic leucocytes images, appliedcomputer image processing and recognition technology, made full use of information incolor images, a method is proposed based on HSI color space to complete white blood cellmicroscopy image segmentation to extract the nucleus, and then use hue information toextract the whole cell, so to get cytoplasm, effectively combined color and regioninformation. Blood cells feature extraction isolated cells from the complex background,analyzed and classified them to extract characteristics. Then selected geometry,morphological features of the image, designed BP neural network classifier for classification.In leukocyte classification and identification stage, in order to solve the problem ofcomplex type of blood cell images that is difficult to identify their characteristics of high dimension, category, to improve recognition accuracy, the main research design of the BPneural network classifier, and achieved satisfactory results.This paper proposed image processing algorithms to recognize five types of whiteblood cells in peripheral blood automatically. Then, a variety of features are extracted fromthe segmented regions. Feature selection method was then used to prioritize importantfeatures and reduce the number of inputs. Finally，networks were trained and evaluatedusing data from segmented leucocytes.