多目视觉三维重建系统在中医舌象诊断中的应用研究

Research on three-dimensional reconstruction of tongue based on multi-vision

目的中医对舌的观察是多方位、立体的。探讨利用多目视觉的三维重建系统对舌象进行信息提取,以补充传统舌像仪的不足。方法搭建快速立体成像系统,在模拟自然光源下,同时采集多幅舌图;采用张氏标定法,进行3D重建,生成立体舌图。选择门诊患者89例,其中男性患者37例,女性患者52例;年龄25~70岁,平均年龄50岁;体质量指数(BMI)16~26 kg/m^2,平均BMI 23 kg/m^2。采用试验设计系统(A组)和对照设备(圣美孚科技有限公司中医综合诊断系统,B组)分别采集患者舌部图像,分析舌色、舌形并比较。结果试验设计采集系统中,相机焦距与标称焦距值之间的误差都很低,最大误差为3.837%,最小误差为0.107%;相机外标定光心间距与实际光心间距的误差在1.731%~3.793%,标定误差很小,标定精度符合实际需求。舌体三维重建图像舌体细节清晰,三维重建效果良好。与B组舌色比较,A组对舌淡白、舌淡红、舌暗红的分析符合率明显高于B组(91.30%vs 86.95%,95.65%vs 95.60%,97.30%vs 94.59%),舌红、舌绛A组与B组符合率一致;A组准确率高达96.85%。舌形判定,A组正常、胖、瘦符合率高于B组(92.9%vs 85.7%,94.1%vs 91.2%,92.6%vs 92.6%)。结论多目视觉三维重建系统是一种有效的舌象分析方法,它可以有效地提高自动判别的正确性,同时对一些舌图特征进行有效的空间量化,可为以后的舌诊仪开发提供理论及临床支持。

Objective To extract tongue image information by multi-vision three-dimensional(3 D) reconstruction technology, and supplement disadvantages of traditional tongue imager, because traditional Chinese medicine observes the tongue in multiple directions and three-dimensionally. Methods The fast stereo imaging system was builded to collect multiple tongue images at the same time under the simulated natural light source. The Zhang calibration method was used for 3 D reconstruction to generate 3 D tongue image. A total of 89 outpatients were enrolled, which included 37 males and 52 females, aged 25-70 years old with mean age of 50 years old;body mass index(BMI) was 16-26 kg/m^2 with mean BMI of 23 kg/m^2. The experimental design system(group A) and control equipment(comprehensive diagnosis system of traditional Chinese medicine, group B) were used to collect images of tongue, and tongue color and shape were analyzed and compared. Results In experimental design acquisition system, the error between camera focal length and nominal focal length was very low, the maximum error was 3.837 %,and the minimum error was 0.107 %. The error between externally calibrated optical center distance of camera and actual optical center distance was 1.731 %-3.793 %, the calibration error was very small, and the accuracy met actual requirements.The 3 D reconstruction image of tongue showed clear details and good results. The analysis coincidence rate of pale tongue,pale red tongue and dark red tongue in group A were significantly higher than those in group B(91.30 % vs 86.95 %, 95.65 %vs 95.60 %, 97.30 % vs 94.59 %). The coincidence rate of red tongue and crimson tongue of group A was consistent with that of group B;the accuracy rate of group A was as high as 96.85 %. For tongue shape, the coincidence rate of normal, fat, and thin in group A was higher than that in group B(92.9 % vs 85.7 %, 94.1 % vs 91.2 %, 92.6 % vs 92.6 %). Conclusion It is demonstrated that multi-vision 3 D reconstruction is an effective tongue image analysis method, which could effectively improve correctness of automatic discrimination. At the same time, it could effectively quantify some tongue image features, which pro-vide theoretical and clinical support for the future development of tongue diagnosis instruments.