基于非均匀采样和方向扩散的指纹图像压缩感知

Compressive fingerprint sensing via nonuniform sampling and directional diffusion

提出了一种新的指纹图像压缩感知方法,采样点比传统压缩感知(CS)方法更少。提出了一种交替使用高倍采样块和低倍采样块的非均匀采样方案。首先,根据从指纹图像训练得到的冗余字典,利用正交匹配追踪(OMP)算法恢复初始图像。然后,根据方向扩散法用高质量的低倍采样块来提升高倍采样块的质量。实验结果表明,该方法显著优于两种现有的知名压缩感知方法。

This paper presents a novel compressive sensing （CS） of fingerprint images, in which less samples are required than the traditional fingerprint sensing methods. It proposes a nonuniform sampling scheme by interleaving high-rate sampling blocks and low-rate sampling blocks in the quincunx manner. The initial image is recovered by the orthogonal matching pursuit （OMP） algorithm using a redundant dictionary trained from fingerprint images. Then, a directional diffusion method is applied to boost the quality of low sampling-rate blocks from high sampling-rate blocks. The experimental results show that the proposed method significantly outperforms two well-known existing compressive sensing methods.