Feature points matching for face reconstruction based on the window unique property of pseudo-random coded image

Abstract

Structured light vision systems have been successfully used for accurate measurement of the 3D surfaces of an object, in which a pseudo-random coded structured light image pattern is projected onto the target object through a projector, and the coded information image produced by its surface is captured by a CCD camera in order to recover its 3D surfaces. In this kind of computer vision technology, 3D face reconstruction is a hot research topic. This paper presents a method for feature points matching used in 3D reconstruction. In this method, the feature points can be identified exclusively taking advantage of the window unique property of a pseudo-random array. Thus, the matching problem can be solved by finding the correspondence between 2D coordinates of feature points in the pixel image and those in the code of the projected template. Then, the 3D reconstruction can be carried out with only a single image with the benefit of easy operation and simple calculation. An experiment for 3D face reconstruction with simulated data is given. The performances show that this method has high matching precision for object matching of feature points.