图像栈的特征提取以及在线虫分类中的应用

发布时间：2019-02-12 19:14

【摘要】：三维多焦距图像栈中的有效信息分布在不同的图像层上,其特征提取及分类跟普通二维图像有很大区别。针对图像栈的分类问题,提出了一种基于多方向图像融合的多线性特征提取和分类方法。首先,通过图像融合获取三维图像栈沿多个正交方向的融合图像,并从中提取特征;然后,通过典型相关分析(canonical correlation analysis,CCA)方法,将不同方向融合图像提取的特征进行融合并抽取组合的典型相关特征用于图像分类;其次,由于图像栈数据包含了样本、类别和方向等多个影响分类的因素,因此将多方向图像融合方法嵌入到多线性分析中,综合考虑多个因素交互作用时对图像栈分类的影响。本文提出的方法在线虫图像栈数据上进行了实验,识别率达到了97.0%,实验结果表明该方法具有较高的准确性。
[Abstract]:The effective information in the 3D multi-focal image stack is distributed on different image layers, and its feature extraction and classification are quite different from those in general two-dimensional images. To solve the problem of image stack classification, a multi-linear feature extraction and classification method based on multi-directional image fusion is proposed. Firstly, the fusion images along several orthogonal directions are obtained by image fusion, and the features are extracted from them. Then, by using the canonical correlation analysis (canonical correlation analysis,CCA) method, the features extracted from different directions fusion images are fused and the combined canonical correlation features are extracted for image classification. Secondly, because the image stack data includes several factors, such as sample, category and direction, so the multi-direction image fusion method is embedded into the multi-linear analysis, and the influence of the multi-factor interaction on the image stack classification is considered comprehensively. The method proposed in this paper is tested on nematode image stack data and the recognition rate reaches 97.0. The experimental results show that the method has high accuracy.
【作者单位】：湖南大学电气与信息工程学院;
【基金】：国家自然科学基金(61301254) 湖南省自然科学基金(14JJ3069)资助项目
【分类号】：TP391.41

【相似文献】