基于稀疏编码的脑脊液图像快速识别模型

Fast recognition model for cerebrospinal fluid images based on sparse coding

考虑到采用传统的图像分割算法很难准确地分割脑脊液(CSF)细胞图像,提出了一种基于稀疏编码的脑脊液图像快速识别模型。该模型首先利用稀疏编码提取图像中的局部特征以及特征描述子,然后将特征描述子转换成线性空间金字塔匹配(SPM)结构,最后将计算结果输入到线性支持向量机(SVM)中进行训练和预测。对脑脊液细胞图像做了异常识别和分类测试,其中异常识别准确率达到了89.4±0.9%,且对每张760×570的图像平均识别时间只需1.3 s,由此可以表明所提出的模型能够有效快速地区分脑脊液细胞是否异常。

Considering the traditional image segmentation algorithm was difficult to segment cerebrospinal fluid cell images accurately, a fast recognition model based on sparse coding for cerebrospinal fluid cell images was presented in this paper. First in this model local features and feature descriptors from the image were extracted by sparse coding. Then the feature descriptors were transformed into linear Spatial Pyramid Matching （SPM） structure. Finally, the calculated result was input into the linear Support Vector Machine （SVM） for training and prediction. In this paper, a test was made for recognizing abnormal cerebrospinal fluid cell images and classification, and the abnormal recognition accuracy rate of the experimental results was up to 89.4：1： O. 9%, and the average recognition time of each 760- 570 image is just 1.3 seconds. Therefore, the presented model can effectively and quickly distinguish normal and abnormal eerebrospinal fluid cell images.