用于面诊的人脸全景图像拼接算法

采用一种基于人脸特征的柱面投影方法，将人的头部近似看做一个圆柱体，有效地解决了在采集过程中因面部角度所引起的视觉不一致性；利用SIFT特征匹配算法提取两幅图像的特征向量，并通过RANSAC匹配优化算法消除错误的匹配，实现图像的配准；接着采用渐入渐出的融合算法，使图像间实现平滑的过渡，消除拼接缝隙。实验结果表明，本研究使用的算法能够快速、有效地生成人脸全景图像，为后续中医面诊奠定了基础。

Face panoramic image mosaic algorithm for face diagnosis

A cylindrical projection method based on facial features is used to approximate the human head as a cylinder, which effectively solves the problem of visual inconsistency caused by different facial angles during image acquisition. SIFT feature matching algorithm is used to extract the feature vectors of two images, and the optimized RANSAC matching algorithm is used to eliminate false matches, thereby realizing image registration. Then fade-in and fade-out fusion algorithm is applied to smooth image transitions and eliminate stitching gaps. The experimental results show that the proposed algorithm can generate the panoramic image of face quickly and effectively, which lays a foundation for the follow-up face diagnosis in traditional Chinese medicine.