基于纹理特征和核Fisher的红细胞识别算法

为解决正常红细胞的计算机自动识别记数问题,提出了一种红细胞识别分类计数算法。首先基于计算机图像处理技术,利用二值图的拓扑特性实现单个细胞的定位。然后根据定位信息提取单个细胞灰度图像数据,计算该细胞图像经小波变换后低频系数的灰度-基元共生矩阵,并提取能量、熵等8个特征作为特征矢量,利用核F isher判别实现对红细胞的识别和计数。实验结果表明该算法具有较高的分类识别正确率,可用于与红细胞形态变化相关疾病的辅助诊断。

Identification algorithm for red corpuscles of blood based on texture feature and kernel-Fisher

An identifying and classifying algorithm was proposed to resolve the problem of counting and distinguishing normal erythrocyte automatically.Firstly,based on the computer image processing,single cell is localized by applying the topology character of Bi-value graph.Then,according to the former orientation,picking up the gray data of single cell,calculating the gray-primitive intergrowth matrix of the low-band coefficients,the cell images were obtained by wavelet transform.At the same time,some classical features, such as energy,entropy were extracted as the feature vectors,distinguished counting are actualized by using the nuclear-Fisher function.Experimental results showed that the proposed algorithm was more approximate than other algorithms,and thus can be applied in the assistant diagnose of diseases that were correlative with the modal change of erythrocyte form.