基于边界支持向量的白细胞检出新方法

New method for white blood cell detection based on boundary support vectors

白细胞检出是血液显微图像白细胞自动识别系统中最基本和最关键的一个环节,其准确性和稳定性直接影响到整个系统的识别准确率。提出了一种基于边界支持向量的白细胞检出新方法。通过引入支持向量回归,在拟合直方图的同时得到稀疏的边界支持向量,再在这些有限的边界支持向量中直接筛选出所需阈值。该方法适用于彩色显微图像分割,能有效地克服光照、染色等客观因素的干扰,具有分割效果优、计算效率高、参数设置简便等优点,有利于后续特征抽取与分类计数。实验结果表明了本文方法的良好性能。

White blood cell detection is one of the most fundamental and key steps in the automatic recognition system of white blood cells in microscopic blood images. Its accuracy and stability greatly affect the recognition accuracy of the whole system. In this paper, a new method for white blood cell detection based on boundary support vectors is proposed. Support vector regression （SVR） is introduced, and sparse boundary support vectors are obtained while fitting the 1D histogram by SVR, and then so-needed threshold values are directly sifted from these limited support vectors. The proposed method is applicable to color microscopic images. It can effectively reduce the influence brought by illumination and staining. It also has the advantages, such as good segmentation performance, high computing efficiency and easy parameter setting, which are helpful for the following feature extraction and classification and for the recognition accuracy improvement of the whole system. Experimental results demonstrate its good performances.