【作者】 王晓东；

【导师】 李传江；

【作者基本信息】 哈尔滨工业大学 ， 控制科学与工程（专业学位）， 2016， 硕士

【摘要】 序列图像的超分辨率重建技术是指对同一景物进行带有亚像素位移的多次拍摄,利用图像之间的互补信息重建出一幅空间分辨率大大提高的图像的过程。图像超分辨率重建对于提高图像的分辨率具有重要作用,并已成功应用于遥感图像领域。SP-20000C-PMCL工业相机由于自身极佳的性能,有望在作为探测器应用到遥感卫星。本文在调研当前图像超分辨率重建方法研究现状的基础上,主要进行了以下工作:首先,为模拟遥感图像的成像过程,本文设计了SP-20000C-PMCL工业相机观测系统。系统由硬件和软件部分组成,其中,硬件部分主要由MEADE LX200望远镜,SP-20000C-PMCL工业相机,图像采集板卡,计算机四部分构成;软件部分主要包括用VS2010开发的SP-20000C-PMCL工业相机的图像控制软件、图像采集软件、图像参数存储与读取软件、插帧软件等。实现SP-20000C-PMCL工业相机对图像的高速采集与实时控制,图像控制参数的的存储与读取。其次,为得到图像超分辨率重建的良好结果,对各种图像配准算法,如光流法、频域法、空域法、特征点匹配法等进行研究。通过对lena原图和加噪后图像进行图像旋转,平移操作,然后用配准算法分别估计位移量,比较不同配准算法的准确度和鲁棒性。通过实验可以得出,在图像只存在平移操作时,光流法具有较高的准确度和鲁棒性。在同时具有旋转和平移变换时,KEREN算法精度较高,且对噪声具有很强的抑制效果,并且对配准误差原因进行了分析。最后,对各种图像超分辨率重建方法,如频域方法、投影方法、总变分法、自适应方法等进行研究,用SP-20000C-PMCL工业相机所得图像作为序列低分辨率图像,进行图像超分辨率重建操作。并从主观和客观上分别对图像重建结果进行评价。评价结果表现出了主观和客观评价结果一致性。更多还原

【Abstract】 Super-resolution reconstructionis a method using a sequence of low-resolution images with complementary information to generate a high-resolution image. Image super-resolution reconstruction plays an important role in improving the image resolution, and has been successfully applied in the field of remote sensing images. SP-20000C-PMCL industrial camera is expected to be as a detector of remote sensing satellites for its great performance. I conducted some experiences based on research of image super-resolution reconstruction methods as follows:First of all, this paper designed the SP-20000C-PMCL industrial camera observation system to simulate the imaging process of remote sensing images. The system consists of hardware part and software part, and the hardware part is mainly composed of a MEADE LX200 telescope, a SP-20000C-PMCL industrial camera, a data acquisition card and a computer. Also, the software part mainly includes an image control software, an image acquisition software,an image parameter storage-reading software and an adding frames software of the camera, which are all developed by VS2010. So we can control the camera in real time and the camera has a high data transmission speed.Then, this paper studied a variety of image registration algorithms, including the optical flow method, the frequency domain method,the spatial domain methods, the feature matching methods. Rotate and translate the lena image and adding-noised lena image and record the registration parameter. Comparing it with the estimated parameters using the image registration algorithms. The experiment result shows that the optical flow method and the keren method have better accuracy and robustness. And the paper analyses the source of the registration error.After selecting a effective registration algorithm, we conduct the experiment of image super-resolution reconstruction.Super-resolution reconstruction methods include the PG method, the IBP method, the POCS method, the TV method and the adaptive normalized convolution method. The sequence of images grabbed by the industrial camera are used as the experimental material. The images generated by the experiment are evaluated by subjective and objective methods. The result of the assessment shows that the effectiveness of the two evaluations.At the end, the paper sums up the experimental research and get a general conclusion. Meanwhile, the paper describes the limitation and insufficiency of the research and prospectsthe direction of future studies.更多还原