The purpose of this paper is to describe the development and performance tests of a contact-free fingerprint sensor, TrueFinger3D (TF3D). This contactless fingerprint sensor is designed to be perfectly interoperable with fingerprint image data captured with contact-based sensors or ink pads. This is achieved by acquiring a 3D dataset of the fingertip together with the image of the papillary lines. Based on the 3D data, the papillary lines image can be processed to compensate perspective foreshortening or even emulate deformation effects caused with contact-based sensors. The 3D measurement mechanism and the image processing are described in detail. The resulting fingerprint images taken by the contactless sensor are then matched with images taken by regular contact-based fingerprint readers at different force levels. The comparison shows that the geometric distortion of our contactless sensor TF3D is comparable to that of contact-based sensors deployed under regular conditions. Our test also shows that contact-based sensors operated under irregular or strong force conditions suffer from a substantial performance degradation, not seen with the contactless sensor TF3D, which has perfect reproducibility. The results also indicate perfect interoperability of the TF3D with any contact-based data and should therefore entitle the sensor to a certification for governmental use. 33 Figure 3: Sample green and blue channel image of the scanner. Please note the grey scale data of the camera channels has been matched to a green, resp. blue color scale. fringe distance is used as a carrier frequency to separate the ideal pattern and the real deformation of the sample. The special challenge for the phase calculation in the finger scanner application is the bad image quality compared with typical interferometric patterns. In Figure 3 it can be seen that the blue camera channel does not detect the fringe projection only: the fringes are disturbed by the papillary line patterns as well as from crosstalk between the camera's colour channels. The visibility of the fringe pattern is increased in a first analysis step by a special combination of the images from the blue and the green camera channel. The interruption of one of the fringes in the middle of the image is caused by an artificial structure in the pattern generator. We use that to detect the zero height reference needed to calculate the real 3D topography of the finger scanned (see section 2.2). Caused by the grazing projection and depending on the actual position of the free floating finger during the measurement, that fringe interruption is detected on varying camera pixels. But it has always the same fringe number in the optical system. The phase image is calculated by a fringe tracking (FTR) method [KBVE93], with some adaption to the special fringe image types detected by the contactless finger print sensor. A 1D Fourier analysis is used in advance of the FTR to check the fringe direction and fringe density. Images not fullfilling the system-inherent limits (e.g. empty measurements, unusual shapes) are sorted out and the relating measurement must be repeated. The following FTR looks for the dark fringes (minima of the intensity) using a first order derivation across the main fringe direction [Sny80]. That results initially in a collection of single data points each representing a local intensity minimum. The second step is the most important - and most difficult one in the phase determination: a highly specialized tracking algorithm connects the single points to fringes giving all points of the same fringe the same order number. Remaining data points with no connection to a fringe are deleted and lead to an area where no phase and finally no fingerprint can be reconstructed. The last step of the phase determination is the calculation of the phase value at each image point. For that we simply use all numbered data points in one camera row (in case of mainly vertical fringes), set the phase difference between two succeeding orders to $2π$and calculate the phase of all image points in-between by a spline interpolation. For a faster processing of the following topography reconstruction we set the zero order to the 36 camera system has been calibrated in a preparatory step and the corresponding intrinsic and extrinsic parameters of the projection and camera system have been considered for the reconstruction process, e.g. the measured surface point coordinates are metrically correct within the accuracy limits of the sensor. Interoperability of sensor devices is very important, e.g. the images shall be suitable for existing minutiae extraction algorithms that were designed to work with fingerprint images acquired with contact-based systems with capacitive sensors or FTIR sensors. A comparison of fingerprint images usually relies on the relative location and orientation of features, such as ridge endings, bifurcations islands, core points or deltas. As the images produced by contact-based devices depict planar finger surfaces all measurements such as minutiae positions and orientations are determined within two-dimensional images. The perspective projection of the fingerprint introduced in the camera based acquisition yields distortions by means of decreased distances between feature points depending on the distance of the observed surface portion. Regarding the shape of a fingertip we can clearly observe that the density of ridges increases from the image centre towards the image border. An ideal solution for this problem would require a projection method which projects the curved finger surface into a plane while preserving both, lengths and angles. A similar problem is well known in the area of map projection, e.g. to create world maps. From a theoretical point of view the situation is quite clear: All map projection methods