Super-resolution Image Reconstruction Based On Nonsubsampled Contourlet Transform

Super-resolution image reconstruction refers to obtain a high-resolution imagefrom a sequence of low-resolution images, these images which must be of the samescene with complementary information are noisy, blurred and downsampled. Thetechnique doesn’t need to improve the existing hardware devices, it just uses the signalprocessing to achieve a high-resolution image. So this technique has wide applicationprospects in remote-sensing, military detection, business, astronnmy, securitymonitoring and other fields.Firstly, this thesis reviews the research development history of the field inSuper-resolution image reconstruction. Then, relevant issues and main technologies ofSuper-resolution image reconstruction are introduced, and the concept of theNonsubsampled Contourlet Transform(NSCT) is also mentioned. Based on themulti-scaled and multi-directional properties of NSCT in signal processing, a newimage Super-resolution reconstruction algorithm is proposed.First sub-pixel image registration is applied. Here the Lucas-Kanade optical flowmethod is combined with Gaussian Pyramid. Then a fused image can be obtained fromthe registered images by performing NSCT and the inverse NSCT. After the lowfrequency and high frequency of the fused image are interpolated, a high-resolutionimage can be obtained by performing the inverse NSCT. The experimental results showthat the proposed algorithm can fuse the complementary information among the givenimages and outperforms the traditional algorithms in terms of both visual quality andobjective evaluation criteria.