基于视觉注意与迁移自学习稀疏表示的血液白细胞分割研究

项目名称： 基于视觉注意与迁移自学习稀疏表示的血液白细胞分割研究

项目编号： No.61772254

项目类型： 面上项目

立项/批准年度： 2018

项目学科： 计算机科学学科

项目作者： 李佐勇

作者单位： 闽江学院

项目金额： 16万元

中文摘要： 白细胞是诊断白血病等血液疾病的重要指标，白细胞分割是血细胞图像自动分析的关键环节。白细胞颜色、形态差异以及细胞质的低对比度等给白细胞分割带来了极大的困难。本项目以提高标准和快速制备下白细胞细胞核分割性能为切入点，探索基于视觉注意的细胞核分割框架。首先，研究基于图像直方图分析的制备方式判断和图像预处理以增强细胞核对比度；接着，在自底向上视觉注意模型中研究基于区域全局对比度和颜色紧致性的显著性度量，用于分割细胞核；最后，在自顶向下视觉注意模型中研究结合先验知识的染色杂质剔除，实现细胞核的最终分割。此项研究将有效地提高标准和快速制备下白细胞细胞核分割性能，为血液白细胞分割和血细胞图像自动分析仪的研发奠定坚实的基础，具有良好的应用前景；同时可以促进相关理论与方法的发展。

中文关键词： 视觉注意模型；稀疏表示；迁移自学习；标签传播；白细胞分割

英文摘要： White blood cell (Leukocyte) is an important indicator of the diagnosis on blood diseases such as leukemia. Leukocyte segmentation is a crucial step of automatic analysis on blood cell image. Leukocyte’s differences on color and morphology as well as the cytoplasm’s low contrast bring great difficulties to leukocyte segmentation. This project aims to improve the segmentation performance of leukocyte’s nucleus under standard preparation and rapid preparation, and to explore a novel segmentation framework based on visual attention mechanism. First, image histogram is analyzed to judge the way of blood smear preparation, and image preprocessing is conducted to enhance the contrast of nucleus. Then, the saliency measure based on regional global contrast and color compactness is used for the nuclear segmentation in the bottom-up visual attention model. Finally, in the top-down visual attention model, prior knowledge is used to remove contaminant for achieving final nuclear segmentation. This research will effectively improve the segmentation performance of leukocyte’s nucleus under standard preparation and rapid preparation, which lays a solid foundation for developing automatic segmentation of leukocyte and automatic blood cell image analyzer with a good application prospects. At the same time, it can promote the development of the theory and methods relevant to this project.

英文关键词： Visual attention model;Sparse representation;Transfer self-learning;Label propagation;White blood cell (Leukocyte) segmentation