【作者】 王志鹏；

【导师】 张桂戌；

【作者基本信息】 华东师范大学 ， 计算机软件与理论， 2007， 硕士

【摘要】 图像复原是当前计算机图形学、计算机视觉和图像处理领域的研究热点之一。它在艺术品的修复、网络数据传输、计算机动画、影视特技、虚拟现实等方面具有广泛的应用前景。图像修补是图像复原研究中的一个重要内容，它的目的是根据图像现有的信息来自动恢复丢失的信息。虽然图像修补基本思想十分简单，但是许多的图像修补算法都十分复杂，而且难于实现。快速行进算法与水平集法(level set)相结合进行曲线进化是一种高效曲线进化算法，算法的时间复杂度是O(NlbN)。Kim等提出了另一种水平集的曲线进化算法—分组行进算法，算法的时间复杂度是O(N)。受其启发，本文将分组行进算法与图像修补算法理论相结合，提出了一种基于分组行进算法的图像修补技术，完成的主要工作包括下面三个方面：●提出了一种基于分组行进算法的非纹理图像修补算法。该算法适合于修补非纹理图像上面的线状区域。与其他算法相比，本文算法在大幅度缩短修补时间的同时仍然能够保持较好的修补效果。●提出了一种基于分组行进算法的纹理图像修补算法。该算法适合于修补包含纹理和结构信息的复杂图像，并且在物体移除方面取得了较好的效果。对于包含纹理和结构信息的复杂图像，本文将基于样本的纹理合成理论与分组行进算法理论相结合，提出一种基于分组行进算法的纹理图像修补算法。实验表明，本文算法能够较好的f冬播图像的纹理信息和结构信息。●引入分组行进算法对窄带(侍修补区域边缘)进行水平集曲线演化，节省了窄带演化的时间，提高了图像修补的效率。为了验证算法的有效性，本文实验如下：非纹理图像线状区域修补、物体移除和纹理图像修补，同时与其他算法的修补结果进行比较。通过比较，发现本文算法在大幅度提高修补速度的同时，仍能保持较好的修补效果。更多还原

【Abstract】 Image restoration is one of the important research hotspots in computer graphics, computer vision and image processing. It has great potential in many applications, especially in restoration of art, fast data transferring on the Internet, computer automation, special effects, virtual reality and so on.Image inpainting is an important research topic in the area of image restoration; its objective is to restore the lost information according to around image information. Although the inpainting basics are straightforward, most inpainting techniques published in the literature are complex to understand and implement. Fast marching method (FMM) is an efficient algorithm for level set applications whose total computation cost is O(NlbN). Kim presented a more efficient algorithm called group marching method (GMM) with the complexity of O(N). Motivated by his work, we propose a new technique for image inpainting based on GMM.The work includes important aspects as follows:A GMM based non-texture inpainting technique is presented. This technique is especially useful for inpainting linear regions on non-texture images. Compared to other techniques, the technique we proposed is much faster while preserving almost the same inpaint result.A GMM based texture inpainting technique is presented. This technique is good at inpainting images contain both texture and structure information, and can also be used for object removal. For texture images inpainting, we introduce GMM for narrow band evolution. It seems that our technique can propagate original image’s structure and texture infomation very well.We introduce GMM for narrow band evolution. Since it takes much less time on narrow band evolution, our technique is more efficient.Three experiments using our algorithm are illustrated, that is, linear regions inpainting on non-texture images, object removal, and texture image inpainting. It seems that our technique is much faster than other inpaint techniques while preserving almost the same result.更多还原