针对光纤束耦合成像系统的特点, 对存在亚像元量级空间错位的图像进行超分辨率重建研究。研究分为图像配准与融合两个步骤。图像配准时, 用经典的 Keren配准算法对两幅图像进行亚像元级别配准。图像融合时结合双三次插值对配准后两幅图像进行 Mallat小波分解与重构, 从而实现图像超分辨率重建。实验结果表明, 简化后的配准算法配准精度达到了 0.01像素, 而经过插值与融合重建出来的图像空间分辨率提高了一倍, 并且保留了丰富的细节信息。

According to the characteristics of fiber bundle coupling imaging system, a research on super-resolution reconstruction of images which has a sub-pixel dislocation was conducted. This research consists of two steps, image registration and image fusion. Classical Keren algorithm, whose registration accuracy was up to sub-pixel-level, was proposed for image registration to register two images. Then, in the stage of image fusion, bi-cubic interpolation is taken to combine with wavelet fusion where Mallat algorithm was put forward for image decomposition and reconstruction. Super-resolution reconstruction of images is realized by such steps. Simulation results show that the simplified algorithm achieved a high registration accuracy of 0.01 pixel, and image reconstructed by interpolation and wavelet-fusion effectively gains a doubled-spatial-resolution with more details.