基于小波变换的真菌隐球酵母菌图像参数压缩

Compress the Image Feature of Cryptococcus neoformans Based on the Wavelet Transformation

提出用小波Haar算法对真菌隐球酵母菌图像参数进行压缩.该方案首先按照行优先的原则,将每个酵母菌图像的特征参数由一维转换为二维;然后利用小波方法将二维参数进行压缩.实验利用机器学习工具WE-KA,选取52个训练集图像以及26个新图像,采用"10舍1"及"10份10轮"交叉验证方法,建立多个预测模型.实验表明小波方案大大节约了运行时间,而在识别变异病原体形态的问题上,几乎与数据驱动特征参数选择方法同样有效.同时因为小波算法的可逆性,特征参数还可以恢复到初始特征参数集合.

This paper presents the wavelet transformation model to compress the image feature of Cryptococcus neoforrnans. Firstly the model is built based on the line wise rule to transform the features of one cell from one dimension to two dimensions. Secondly compress the image features using wavelet transformation and build the predication model. At last, evaluate the model comparing with the data driven feature selection methods. The experiments use WEKA machine learning tool, using 52 cell images as training data set and 26 cell images as the test set, ＂leave one out＂ and ＂ten folds ten rounds＂ cross validation are utilized. The results show that wavelet transformation, with much less running time, is almost as powerful as data driven feature selection to identify the pathogen yeast. For the inverse of the wavelet transformation, this method can also be used to recoustruct the original set of the feature.