基于DeepLab改进网络下的舌图像分割算法  

Improved Tongue Image Segmentation Algorithm Based on Deeplab v2 Network and OSIM Network

作　　者：徐武[1] 陈盈君 汤弘毅 杨昊东 XU Wu;CHEN Ying-jun;TANG Hong-yi;YANG Hao-dong(School of Electrical Information Engineering,Yunnan Minzu University,Kunming Yunnan 650000,China)

机构地区：[1]云南民族大学电气信息工程学院,云南昆明650000

出　　处：《计算机仿真》2022年第6期210-213,489,共5页Computer Simulation

基　　金：国家自然科学基金(U1802271)。

摘　　要：舌图像是中医判断身体的重要指标。在舌图像处理过程中,分割结果会受到嘴唇、下巴、脸部等固定干扰因素的影响,以及不可预测的无关噪音遮挡舌体时不固定干扰因素的影响。在精确地分割出完整的舌图像方面,传统分割方法有很大的局限性,针对这个问题现提出基于神经网络DeepLab v2下的舌图像分割改进算法。首先对舌体进行定位,引入带孔卷积算子,通过设置池化层参数获得更好的边缘信息;然后采用生成对抗网络OSIM对无关噪音进行去噪处理,分割出完整的舌体图像;最后通过基于孔洞的空间金字塔融合多尺度的特征信息,使分割结果更具鲁棒性。实验结果表明,上述方法提升了分割精度,显著增强了抗噪能力,并降低在分析过程中固定干扰因素的影响,得到精确分割的舌图像。Tongue image is an important index to judge the body in Traditional Chinnese Medicine. In the process of tongue image processing, the segmentation results will be affected by fixed interference factors such as lips, chin, face, etc., and when the tongue body is covered by unpredictable irrelevant noise, the processing results will produce deviation. The speed and accuracy of traditional segmentation methods are limited. In the paper, an improved tongue image segmentation algorithm based on neural network deeplab v2 is proposed. Firstly, the tongue body is located and convolution with holes was proposed, and better edge information was obtained by setting pool layer parameters;Secondly, the noise of irrelevant noise was denoised by generating countermeasure network OSIM Network, and the complete tongue image was segmented;Finally, multi-scale feature information was fused by spatial pyramid based on holes to make the segmentation result more robust. Experimental results show that this method improves the segmentation accuracy, significantly enhances the anti noise ability, and reduces the influence of fixed interference factors in the analysis process, so as to achieve the purpose of accurate segmentation.

关 键 词：舌图像 神经网络 生成对抗网络 空间金字塔 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]