不同光谱照明条件下的牙齿图像分割

Teeth image segmentation under various spectrum illumination conditions

对不同光谱照明条件下的牙齿图像进行分割有助于将在不同光谱照明条件下发生位置变动的牙齿对齐,从而实现对牙齿色度坐标的精确测量,为口腔修复术中的牙齿颜色匹配打下坚实的基础。然而,分割模型在不同光谱照明条件下表现出的性能并不一致。研究结果表明,在白光照明条件下的高对比度图像中,采用人工标记的牙齿多光谱图像来训练的用于分割单个牙齿实例图像的深度神经网络Mask R⁃CNN能够获得良好的分割结果,但在红、蓝或紫色光照明条件下的部分低对比度的牙齿图像中模型无法正确分割。为了解决这一问题,文中使用了一种现有的光谱重建方法来增强训练数据。实验结果表明,使用该数据增强方法的模型在不同光谱照明下的平均分割精度从对比模型的84.9%提高至89.2%,该数据增强方法显著提高了牙齿分割模型在不同光谱照明条件下的性能,并增强了模型的鲁棒性和泛化能力。

Segmentation of tooth images under various spectral illumination conditions helps to align teeth that undergo positional changes,enabling accurate measurement of tooth chromaticity coordinates and providing a reliable foundation for tooth color matching in restorative dentistry.However,the segmentation model does not consistently perform well under different spectral illumination conditions.The results demonstrate that Mask R⁃CNN,a deep neural network(DNN)trained to segment individual tooth example images using artificially labeled multispectral images of teeth,can achieve good segmentation results in high⁃contrast images under white lighting conditions,but the model struggles to segment correctly in some low⁃contrast dental images under red,blue,or violet lighting conditions.To address this issue,this study utilized an existing spectral reconstruction method to enhance the training data.The experimental results indicate that the average segmentation accuracy of the model,using this data enhancement method under different spectral illumination,improved from 84.9%to 89.2%.This data enhancement method significantly improves the performance of the tooth segmentation model under various spectral illumination conditions,enhancing the robustness and generalization ability of the model.