The skin surface becomes wrinkled and rough due to various internal and external factors. A three-dimensional (3D) analysis of the skin is required to improve skin conditions. Stereophotogrammetry, a noninvasive 3D analysis method, is easy to install and use, but most stereo systems have a fixed baseline and scale. Previous stereo systems are not suitable for observing micro-range skin features. Therefore, we suggest the optimal conditions and methods for the 3D analysis of skin microrelief using a multi-conditioned stereo system. We constructed a nonconvergence model using a mobile device and acquired stereo images under multiscale and multi-baseline conditions. We extracted 3D information of the skin through our process: preprocessing, skin feature extraction, feature matching, and actual depth mapping. We improved the accuracy of the 3D analysis of the skin by using disparity values instead of disparity maps. We compared and analyzed the performances of six local feature detector and descriptor algorithms. In addition, we suggested depth-mapping formulas to estimate the actual depth of the skin microrelief. We confirmed that stereo images with a working distance of 70-75mm and a baseline of 4-8mm are effective for the 3D analysis of skin microrelief. In addition, accelerated KAZE exhibited the best performance for features extraction and stereo matching. Finally, the extracted 3D information was converted to the actual depth, and the performance of the 3D analysis was verified. The proposed system and method that provide texture information are effective for 3D skin disease analysis and evaluation.
Read full abstract