自动指纹识别系统大多是基于细节点匹配的，系统性能依赖于输入指纹质量。输入指纹质量差是目前自动指纹识别系统面临的主要问题。为了提高系统性能，实现对低质量指纹的增强，提出了一种基于多尺度分类字典稀疏表示的指纹增强方法。方法 首先，构建高质量指纹训练样本集，基于高质量训练样本学习得到多尺度分类字典；其次，使用线性对比度拉伸方法对指纹图像进行预增强，得到预增强指纹；然后，在空域对预增强指纹进行分块，基于块内点方向一致性对块质量进行评价和分级；最后，在频域构建基于分类字典稀疏表示的指纹块频谱增强模型，基于块质量分级机制和复合窗口策略，结合频谱扩散，基于多尺度分类字典对块频谱进行增强。结果 在指纹数据库FVC2004上将提出算法与两种传统指纹增强算法进行了对比实验。可视化和量化实验结果均表明，相比于传统指纹增强算法，提出的方法具有更好的鲁棒性，能有效改善低质量输入指纹质量。结论 通过将指纹脊线模式先验引入分类字典学习，为拥有不同方向类别的指纹块分别学习一个更为可靠的字典，使得学习到的分类字典拥有更可靠的脊线模式信息。块质量分级机制和复合窗口策略不仅有助于频谱扩散，改善低质量块的频谱质量，而且使得多尺度分类字典能够成功应用，克服了增强准确性和抗噪性之间的矛盾，使得块增强结果更具稳定性和可靠性，显著提升了低质量指纹图像的增强质量。
Objective Most automatic fingerprint identification systems (AFISs) are based on the minutiae matching. The accuracy and reliability of the minutiae extraction are largely dependent on the quality of input fingerprint image. So, the performance of these AFISs are largely determined by the quality of input fingerprint. In practice, the quality of a fingerprint image may suffer from various impairments, and the image may appear in ridge adhesions, ridge fracture or contrast uneven and so on. In order to improve the performance of the AFIS, the quality of fingerprint image must be enhanced. This paper proposes a novel fingerprint enhancement algorithm using sparse representation by multi-scale classification dictionaries. Method First, we sample the training fingerprints with high quality to build the training set for multi-scale classification dictionaries learning, and the multi-scale classification dictionaries are learned from training set. To make meaningful fingerprint image enhancement, a crucial issue is to obtain an effective prior or constraint. Unlike the generic image, the fingerprint image has a steady and reliable ridge pattern. To obtain an effective prior or constraint, the fingerprint patches orientation are estimated by the weighted linear projection analysis (WLPA) based on the vector set of point gradients. We classify the training samples with the same size into eight groups by their own ridge orientation pattern. Instead of learning simply a dictionary, we learn a classification dictionary for each class with the same size. Second, fingerprints are pre-enhanced by using linear contrast stretching (LCS) method, the pre-enhanced fingerings are obtained. The sparse grey space in fingerprint image is used, and the fingerprint image contrast can be stretched out to cover the whole greyscale space. By doing so, it can completely preserve the gray level information of input fingerprint against loss and as well as achieve better contrast enhancement. It contributes to the subsequent enhancement. Third, fingerprint has a unique natural pattern, and it is very suitable for frequency domain analysis. Generally, a good frequency-domain fingerprint enhancement approach is designed to work on spatial-partitioning and frequency-domain enhancing. So, the fingerprint is partitioned into patches in spatial domain based on non-overlapping window, the orientations of fingerprint patches are estimated by the weighted linear projection analysis and the qualities of the patches are evaluated and classified by the coherence of point orientations. And finally, the fingerprint patches are transformed to frequency domain by the 2-D discrete Fourier transform. The enhancement model of patch spectrum is constructed via sparse representation modeling by using classification dictionaries. The patch spectrum are enhanced based on quality grading scheme and composite strategy by using multi-scale classification dictionaries learning combined with spectra diffusion. The fingerprint patch is enhanced according to its own priority, and the patches with higher quality have been enhanced when the patches with lower quality are enhanced. The multi-scale classification dictionaries learning ensure the reliability of the enhancement. The spectra diffusion is successfully applied with the help of the scheme of quality grading and neighborhood priority and the strategy of composite window, and it can improve the quality of the patch spectra with lower quality. It provides more accurate ridge spectra information for lower quality patches, guarantee the reliability of the ridge spectra for the multi-scale classification dictionaries enhancing. Result The proposed method has been implemented and tested on fingerprint images from FVC2004. Some visual experiments and performance evaluations of the minutiae extractions are illustrated. We compare our method with other state-of-the-art fingerprint enhancement methods and can report that the proposed method is superior for enhancing fingerprint images. The experimental results demonstrate that the fingerprint with low quality can be effectively enhanced by the proposed method. Compared with the traditional fingerprint enhancement algorithms, the proposed method is more robust against noise, and has a prominent effect on low quality fingerprint image. Conclusion By introducing the ridge pattern priori into classification dictionary, a classification dictionary for each class with the same size is learned. The classification dictionaries based on ridge pattern constraint can capture the more reliable ridge pattern prior. Use of classification dictionaries improve the effectiveness of sparse modeling of information in a fingerprint patch. The quality grading scheme and the composite window strategy are employed to assist the multi-scale dictionary to overcome the contradiction between accuracy and anti-noise ability. Furthermore, the combination of composite window and quality evaluation promises to ensure that spectra diffusion is successfully applied. The proposed method significantly improves the quality of input fingerprints with low quality.