动漫图像交互式纹理替换算法的研究与实现

发布时间：2018-01-31 12:32

本文关键词： 动漫图像图像分割语义级纹理推荐纹理合成光照定制　出处：《深圳大学》2017年硕士论文　论文类型：学位论文

【摘要】：伴随网络的普及以及动漫产业自身的高速成长,动漫的市场需求量也相应的加大。然而动漫的制作是一个繁琐复杂的过程,动漫产品的诞生需要长时间的制作周期,这就导致了动漫产业的高投入低产出。因此如何提高动漫产品的产量,关键在于提高动漫的制作技术,缩短制作周期。经过对动漫制作的深入观察和研究发现,虽然每一帧漫画都是人手画出的,每一个细节都需要精心雕刻,但是,动漫图像还是存在一些重复出现的元素,比如沙滩、草坪、织物表面、墙纸、地表、树的表皮等。这些区域的绘制是否可以用其他高效的方法替代?针对这一系列问题,本文提出动漫图像交互式纹理替换的方案,可以在很大程度上节约时间,提高效率。本文围绕这个方案分别对以下三个方面展开研究。首先是图像分割,进行纹理替换的前提是将该区域与其他区域区分开,这就涉及图像处理中的图像分割。为此在对多种主流的图像分割算法的深入分析研究之后,本文借助SLIC超像素的算法理念提出了全新的分割算法。在SLIC超像素生成算法的原理上,对该算法进行了改进和创新,得到更好的实验结果,使得分割效果更契合人类对分割的定义。其次是纹理推荐,在成千上万的图片中寻找合适的纹理本是一件繁琐乏味的过程,纹理推荐可以很好的解决这个问题。本文在原有的纹理特征提取算法上,总结出一系列新的纹理度量值,例如纹理性、同质性、重复性等。此外,我们还建立一个庞大的纹理样图库。并结合这些纹理特征,实现语义级纹理推荐,通过设置纹理特征值,系统就可以检索纹理图库推荐出相应的纹理样图。最后是纹理合成,将纹理样图完美的替换到原图上是一个关键的步骤,在综合研究分析了多种纹理合成算法的各方面性能,本文选择了Graphcut纹理合成算法,并对该算法实现并行化改进,并且在原算法的基础上提出了光照定制这一创新点,可以根据输入光照图像来影响合成效果。在时间性和合成效果上,使得Graphcut算法都得到改进。
[Abstract]:With the popularity of the network and the rapid growth of the animation industry itself, the market demand for animation is also increased accordingly. However, the production of animation is a complicated process. The birth of animation products needs a long production cycle, which leads to the animation industry high input and low output. So how to improve the production of animation products, the key is to improve the animation production technology. Shorten the production cycle. Through the animation production in-depth observation and research found that although each frame of cartoon is hand-drawn, every detail needs to be carefully carved, but. There are still some recurring elements in anime images, such as beach, lawn, fabric surface, wallpaper, surface, tree skin and so on. Can these areas be replaced by other efficient methods? Aiming at this series of problems, this paper proposes an interactive texture replacement scheme for animation images, which can save time to a great extent. This paper focuses on the following three aspects: firstly, image segmentation, the premise of texture replacement is to distinguish the region from other regions. This involves image segmentation in image processing. In this paper, a new segmentation algorithm is proposed based on the idea of SLIC super-pixel algorithm. In the principle of SLIC super-pixel generation algorithm, the algorithm is improved and innovated, and better experimental results are obtained. Make the segmentation more in line with the human definition of segmentation. Secondly, texture recommendation, looking for the appropriate texture in thousands of images is a tedious process. Texture recommendation can solve this problem very well. In this paper, a series of new texture measures, such as texture, homogeneity, repeatability and so on, are summarized in the original texture feature extraction algorithm. We also build a huge texture map library, and combine these texture features to realize the semantic level texture recommendation, by setting the texture feature value. The system can retrieve the texture database and recommend the corresponding texture image. Finally, the texture synthesis is the key step to replace the texture image to the original image perfectly. In the comprehensive research and analysis of the various aspects of the performance of texture synthesis algorithms, this paper selected the Graphcut texture synthesis algorithm, and the algorithm to achieve parallelization improvement. On the basis of the original algorithm, the innovation of illumination customization is put forward, which can affect the synthesis effect according to the input illumination image. In terms of timing and composition effect, the Graphcut algorithm is improved.
【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：J218.2;TP391.41

【相似文献】