Fingerprint segmentation and quality map using a combined frequency model

As the main purpose of fingerprint images is to provide a good way of identifying individual persons within a huge set, we need to extract specific properties of them and investigate image regions depending on their quality. In the tool chain of processing fingerprint images one has often a need for a segmentation of the image into foreground e.g. the fingerprint, and the background that contains image parts without any or unusable fingerprint regions. Many methods use only the measurement of local information in a block such as coherence, mean and variance as well as combinations from them. We propose a novel method for segmentation that uses local as well as global information based on different frequency ranges in the Fourier space. This combined frequency model is then used for the segmentation. An evaluation of the method based on ground truth data is provided and compared with state of the art methods. We generate a continuous measurement of fingerprint occurrence within the image denoted as the quality map. This map is used to improve the overall system stability and performance in further steps like matching, enhancing direction field extraction and so on. This paper will focus on the evaluation of fingerprint quality maps and segmentations based on them.