School of Information and Communication Engineering, Harbin Engineering University, Harbin, 150001, China.
Adv Exp Med Biol. 2011;696:255-62. doi: 10.1007/978-1-4419-7046-6_25.
The multi-target tracking in cell image sequences is the main difficulty in cells' locomotion study. Aim to study cells' complexity movement in high-density cells' image, this chapter has proposed a system of segmentation and tracking. The proposed tracking algorithm has combined overlapping and topological constraints with track inactive and active cells, respectively. In order to improve performance of algorithm, size factor has been introduced as a new restriction to quantification criterion of similarity based on Zhang's method. And the distance threshold for transforming segmented image into graph is adjusted on considering the local distribution of cells' district in one image. The improved algorithm has been tested in two different image sequences, which have high or low contrast ration separately. Experimental results show that our approach has improved tracking accuracy from 3% to 9% compared with Zhang's algorithm, especially when cells are in high density and cells' splitting occurred frequently. And the final tracking accuracy can reach 90.24% and 77.08%.
细胞图像序列中的多目标跟踪是细胞运动研究中的主要难点。本研究旨在研究高密度细胞图像中细胞的复杂运动，提出了一种分割和跟踪系统。所提出的跟踪算法分别将重叠和拓扑约束与跟踪活动和不活动细胞相结合。为了提高算法的性能，已引入大小因子作为基于张方法的相似性量化标准的新限制。并且考虑到一个图像中细胞区域的局部分布，调整了将分割图像转换为图的距离阈值。改进的算法已经在具有高或低对比度比的两个不同的图像序列中进行了测试。实验结果表明，与张算法相比，我们的方法将跟踪精度从 3%提高到 9%，特别是在细胞密度高且细胞分裂频繁时。最终的跟踪精度分别可以达到 90.24%和 77.08%。
