Feature Selection And Recognition Of Color Leukocyte Images

Digital image processing and analysis has been widely applied in many medical areas. Advanced image processing and pattern recognition technology, which is used in the sum and sort countings of leukocytes, is one of the important methods in medical-aided diagnosis. The technology partly replacing the labor works and lightening the burden of doctors by effectively reduced subjective influence, highly increases the diagnostic efficiency.Basic study done in this paper is based on the methods of automatic classification of leukocytes in peripheral blood smear, bringing forward a set of basic algorithms including pretreatment, forming and selecting features as well as classification criteria. Most work has been focused on the feature extraction and selection. Satisfying results have been reached by the selected feature parameters acting on leukocytes classification.According to the eyeballing experience of leukocytes classification, combining with the former work in feature extraction, quantificational described morphology parameters, such as shape and size of cell area and branch numbers of nuclear are extracted on the basis of single leukocyte cell segmentation. Following work has been done in color parameters extraction, most of which are statistical color characteristics based on region information. Various algorithms, including the discrete wavelet transform and an effective texture operator, local binary pattern, are used in texture parameter extraction, by which an effective feature subset is formed. Considering the complexity of different parameters, a step-by-step strategy is adopted in feature selection of leukocytes to get optimalized feature parameters. A subset efficient to classification is composed of the features selected by Gene Algorithm which is firstly used in texture features selection and followed by Tabu Search adopted in a secondary selection within the general feature vector sets. Also the capability of our method is compared to that of ther feature selection algorithms.Several neural network classifier models are constructed for leukocytes recognition. With a small quantity of samples, our proposed methods achieve a decent performance of leukocytes classification.