光学学报, 2020, 40 (17): 1710001, 网络出版: 2020-08-24
基于残差块和注意力机制的细胞图像分割方法 下载: 1037次
Cell Image Segmentation Method Based on Residual Block and Attention Mechanism
图像处理 细胞分割 U-Net 残差块 注意力机制 image processing cell segmentation U-Net residual block attention mechanism
摘要
针对相衬显微镜采集的细胞图像具有亮度不均衡且细胞与背景对比度较低的问题,提出一种以U-Net为基本框架,结合残差块和注意力机制的细胞分割模型。首先,利用具有编码器-解码器结构的U-Net对细胞图像进行细胞初始分割;然后,在U-Net中引入残差块,以强化特征的传播能力,提取更多细胞细节信息;最后,利用注意力机制加重细胞区域的权重,降低亮度不均衡、对比度较低对模型的干扰。实验结果表明,与其他模型相比,所提模型在视觉效果和客观评价指标上均有较好的分割效果。
Abstract
Herein, a cell segmentation model is proposed based on the U-Net and combination of residual block and attention mechanism. This model aims at addressing the problems of uneven brightness and low contrast between a cell and background of cell images collected using a phase contrast microscope. First, the U-Net with encoder-decoder structure is used to conduct the initial segmentation on cell images. Hereafter, the residual block is introduced into the U-Net to strengthen the propagation ability of the features and extract more cell-detail information. Finally, the attention mechanism is used to increase the weight of the cell area and reduce the interference of uneven brightness and low contrast on the model. The experimental results show that compared with other models, the proposed model exhibit better segmentation results in visual effects and objective evaluation indicators.
张文秀, 朱振才, 张永合, 王新宇, 丁国鹏. 基于残差块和注意力机制的细胞图像分割方法[J]. 光学学报, 2020, 40(17): 1710001. Wenxiu Zhang, Zhencai Zhu, Yonghe Zhang, Xinyu Wang, Guopeng Ding. Cell Image Segmentation Method Based on Residual Block and Attention Mechanism[J]. Acta Optica Sinica, 2020, 40(17): 1710001.