基于改进LBP特征的白细胞识别

Leukocyte Classification Based On an Enhanced Local Binary Pattern Feature

采用先进的计算机图像处理与分析技术完成白细胞分类计数是辅助诊断血液疾病的重要方法。各种白细胞间的纹理差异较大,纹理是区分白细胞的重要特征之一。局部二进制模式(local binary pattern,LBP)是一种有效的纹理描述算子。本研究提出了一种提取细胞的改进LBP特征用于白细胞分类识别的算法。首先,用小波变换对图像进行分解并重构,获得四幅不同频率的分量图,对其中的低频信号采用离散余弦变换。此后,用可变大小的子窗口对变换后的图像扫描,并根据不同区域赋以权值,获取改进的LBP特征,构成加权直方图。这种特征既能反映细胞局部特征又能反映整体特征。根据测试样本和模板的LBP特征直方图之间的马氏距离构建分类器。根据这种改进LBP特征有效地实现了白细胞的5种分类,得到了令人满意的分类正确率。实验结果表明:本研究提出的算法能有效地区分不同类的白细胞;与其他一些算法相比,提高了分类的精确度。

Classification and counting of leukocyte types, by dint of the advanced computer technology for image processing and analysis, is of great importance due to its crucial role in study of assistant diagnosis of blood diseases. Leukocytes are different in texture which is one of the important features in Cell classification. Local binary pattern （LBP） is an efficient operator because of its excellent capability of description of local texture. Here, a method based on extracting an enhanced LBP feature, for identification of leukocytes is introduced, First discrete wavelet transform （DWT） is adopted to decomposed images into four kinds of frequency subimnges, and then discrete cosine transform （DCT） to the low frequency subimages from which the features arc calculated to increase original data. Then the transformed subimages are scanned with small changeable windows from which improved features are obtained and weighted histograms which effectively express both local and holistic features of the cell areas are constructed according to gray - scale distributions. Finally, Mahalanobis distance between corresponding LBP histograms of the test image and template h used to construct classifiers. With the enhanced LBP feature extracted by this method, the satisfying classification accuracy achieves good effects for dividing leucocytes into five types. The results of experiments show that the proposed approach which is discriminative for leukocytes classification achieves better performance of leukocytes classification than the others methods.