基于形状特征的硅藻显微图像自动识别

Automatic Identification of Diatom Microscopic Images Based on Contour Features

硅藻是一类广泛分布于各类生境的单细胞生物,在许多领域具有广泛的应用,如水质监测、环境调查、石油勘探等,而这些应用都离不开对硅藻的种类鉴定.根据硅藻显微图像的形状特点,提出了对其进行预处理、分割、形状特征提取和分类的自动识别方法.采用了基于累积直方图的双轮廓叠加法的图像分割方法,可以有效抑制光照强度不一致、不平衡的影响,并充分利用显微镜下硅藻图像的轮廓特点,获得较好分割效果.同时对硅藻图像提取几何描述全局特征及形状签名特征,最后采用了基于误差反向传播算法的多层前馈网络(BP网络)进行分类.实验表明,该方法对11种浮游硅藻(包括12类轮廓)的自动识别率达到96.6%.

Diatoms are unicellular microscopic algae found in practically any moist environment.Identification of diatom has applications in many disciplines,including water quality monitoring,environment investigation and oil exploration.In this research,a method based on contour features for automatic identification of digital microscopic images of diatoms is presented.This method includes image preprocessing,image segmentation,contour features extraction and classification.At the stage of image segmentation,the best results were obtained by a double contours surperposition method based on accumulation histogram,which makes full use of the characters of diatom contour and reduces effects of nonuniform illumination.Then we extracted morphometric descriptors global features and shape signature features from diatom,and finally use a BP neural network to effectively classify the diatoms.With classification carried out using a BP neural network we attained 96.6% accuracy from a set of images containing 11 species of diatoms which have 12 kinds of contours.