【作者】 张云；

【导师】 杜军平；

【作者基本信息】 北京邮电大学 ， 计算机科学与技术， 2013， 硕士

【摘要】 本论文对基于空间运动图像的增强和超分辨率重建研究,主要针对空间飞行器定位、对接等情景中的图像处理问题进行研究。针对空间运动图像的特点和应用中出现的问题如边缘模糊、噪声影响以及实时性处理等对算法进行了研究。提出了基于多尺度变换的图像增强和多种不同情况下的超分辨率重建算法,设计与实现了图像增强和重建软件工具台。论文完成的主要工作如下：(1)实现了图像增强算法。对超分辨率重建中使用的图像进行了增强。提出了基于多尺度变换的图像增强方法。利用非下采样轮廓波变换将图像分解为多层高低频图像,对不同的高频变换系数采用自适应的阈值进行处理,分部处理后再逆变换为原图像。(2)实现了单帧图像超分辨率重建算法。针对传统方法中的边缘模糊、纹理损失以及复杂运算等问题,提出了基于边缘提取的图像插值重建方法,对图像分区域采用不同的方法进行处理。采用改进的canny算子进行边缘提取并扩张,对边缘独立重建后与其他区域重建结果融合在一起,最终得到结果图像。(3)实现了多帧图像及视频超分辨率重建算法。提出了基于非局部平均估计的改进重建方法。通过综合利用帧间的局部信息,利用新的权值计算公式提高重建效果,并使用预分类、快速统计等方法,提高了算法的运算速度。实验表明该方法的算法速度能达到处理要求。以此算法为核心实现了视频的超分辨率重建,对滑动窗口内的视频帧依次重建后合成为结果视频。(4)开发了图像增强和重建软件工具,验证了本文提出的算法的有效性。更多还原

【Abstract】 The propose of this thesis is image enhancement and image super resolution reconstruction of the space motion images. The study aims at the image process of the spacecraft’s location and docking. The issue studies the problems in space activities like edge blur, noise and so on. The thesis proposed a new image enhancement method and some new algorithms of super resolution reconstruction in different circumstances based on multi-scale transform. It also developed an experimental platform by Matlab. The main works of this thesis are as follows.1) Image enhancement algorithm. The issue proposed a new image enhancement method based on multi-scale transform. First, decompose the image into different frequency by NSCT. Then handle the high-frequency coefficients by adaptive threshold. The result images can be obtained by inverse transformation.2) Super resolution reconstruction with single frame. To conquer the problem of the traditional methods like edge blur and complex calculations, the issue proposed a new fast reconstruction method based on edge extraction. The method can handle the different areas of the image by different ways. It extracts the edges of the image by improved canny operator and does the independent reconstruction on the edge. Then, combine the reconstruction results of different areas to get the high resolution image.3) Super resolution reconstruction with multi-frame images and video. The traditional methods of multi-frame images rely on accurate motion estimation. The issue proposed an improved reconstruction method based on non local means. This method can improve the images’ quantity by the new weight calculation formula. By pre-classification and fast count, the whole algorithm’s speed can be more quickly than others. The method can also apply to the video reconstruction by setting a sliding window.4) Develop an experiment platform. In the end of this thesis, an experiment platform is developed to verify all the proposed methods. The platform provides us a convenient way to deal with images and evaluate the results.更多还原