Three-Dimensional Wound Reconstruction using Point Descriptors: A Comparative Study

Precise measurements of chronic wound areas are very important for measuring the efficacy of different treatments, since it provides a clear picture of the chosen treatment evolution. The most used techniques still in use for measuring the wound areas involve manual measuring, paper scales and/or acetate tracing, with each health professional possibly measuring the wounds in slightly different ways, specially when dealing with the wound's depth. Thus, it is common to observe inconsistent wound area estimations by different health workers on the same wounds. Moreover, the measuring process requires direct contact with the wound, increasing the contamination and infection risks, besides possibly causing discomfort for the patient. An alternative that has already been proved successful is the use of Structure from Motion (SfM) to recover three-dimensional representations of a wound's surface and estimate that area. In this work, we analyze the effectiveness of several widely used point descriptors (BRIEF, DRINK, FREAK, ORB, SIFT and SURF) in the SfM wound reconstruction task when applied to real images of realistic synthetic wounds made of latex placed on different areas of the body. The results showed that SIFT, SURF and DRINK produced better area estimates. Since the execution time is a factor, we selected DRINK as the standard descriptor for our system, since it is approximately 9.3 and 4.9 times faster than SIFT and SURF, respectively.