Threedimensional measurement is one of the primary interests of various industries such as quality control, documentation of cultural artifacts and medical image processing. In recent years, contrary to the contactbased mechanical methods, active imagebased methods using laser light [1] or white light [2] for recovering the surface of the object, have gained considerable attention because of their noncontact nature. Among these techniques, the phaseshifting digital fringe projection (DFP) is wellestablished due to its advantageous characteristics such as full resolution and high accuracy measurement [3]. Extremely dense point clouds that are obtained from structured light 3D scanners leads to many problems in further processing steps. The processing, modeling and visualizing of huge amount of points is a very hard problem for conventional computers [4].
Most of the simplification methods suggest maintaining details. As a result, the high frequency noise in data are also kept during simplification process which leads to decrease the signal to noise ratio (SNR) in simplified data. Although the goal of the simplification step is to decrease further complexities in processing of point cloud, this simplification process still encounters high computational complexity such as search for neighbor points in threedimensional space, curve fitting and normal extraction of surfaces. All existing simplification methods are applied after point cloud generation in the postprocessing step. So, there is a waste of costs in calculations on the processing of a part of data which is not required to be generated during the threedimensional measurement. Therefore, an algorithm/method that smartly generates the minimum required points that perfectly reconstruct the object can efficiently decrease the postprocessing cost.
A hybrid scanner system is proposed in this paper that prevents the production of unnecessary points during measurement by DFP technique, using the geometric characteristics of the surface that are obtained from Photometric Stereo (PS) technique. The PS method generates surface normal from the object surface. The surface curvature can be obtained for each image pixel using PS normal. The curvature image is classified and assigned to the different level of densities. The density levels are defined in image of the stereo camera in scanner system. So by removing pixels in regular numbers the density of each level is constructed. Hence, the first level of density is the same as maximum resolution of camera. Next levels are equal to 50 percent, 33 percent, 25 percent, and 20 percent of all pixels which are respectively result of sampling every other pixel, sampling one pixel from every two pixels, from every three pixels, and from every four pixels. Though, the question arises here that what is the criteria of determining the curvature intervals which are assigned to density levels. The basis of determining the curvature intervals for point simplification is the distance between simplified points and surface computed from original dense point cloud. This distance is chosen by user. In this paper it is equal to measurement system accuracy. So the unnecessary points are removed based on the curvature obtained via PS before calculation of 3D coordinates using FP technique. The surface normal obtained from PS has low highfrequency noise, so noisy data will not transfer to the simplified points. The simple hardware setup of PS technique provides an efficient tool for simplified measurement of DFP scanner. Also, the extraction and classification of geometric features of the object are performed in twodimensional space with lower complexity in comparison with similar operation in threedimensional space. The addition of PS method only adds the cost of a number of light sources to scanner system and also includes the addition of several images to the process of measurement. But on the other hand, it will reduce calculation time by 50% to 75% and will reduce the volume of data by 50% to 80% depending on the complexity of the geometry of the object. The reduction in density has been done with the assumption of maximum separation of simplified model from the main model with the distance of 0.01 mm and 0.02mm.
The principal idea of the proposed method is to measure surfaces with the minimal required points that preserves the geometry. Contrary to most of the simplification methods, the proposed method performs simplification while measures the surface, so no more postprocessing step for simplification is required. At first, surface normal are calculated by PS technique. Then surface curvatures are computed for each pixel in camera image from normal vectors and classified. Each class represents a point density level. The surface slope is also considered to correct foreshortening effect that is caused by projective geometry. The output of previous step is a 2D simplification guidance map which is used to measure 3D objects surface with DFP technique. All measurement steps by proposed system are as follows:
 computing surface Normal and curvature
 assigning curvature ranges to point density levels
 3D measurement by DFP based on simplification map
