闭合形式抠图算法比较及人像植入工具开发
时间:2018-08-26 19:42 来源:毕业论文 作者:毕业论文 点击:次
| 摘要：目前主流的图像处理工具自动抠图功能存在人像边缘比较复杂情况下，例如抠取细丝毛发部分，不能很准确地将毛发抠取出来。在自然图像抠图算法中闭合形式抠图算法可以获取图像中的边缘细节。因此本文选择三种闭合形式抠图算法：共享抠图算法、广泛采样抠图算法、KNN抠图算法来研究。目前抠图算法的比较存在如下问题：(1)比较实验中没有系统全面地从不同场景情况下进行比较。(2)由于一幅图像中可能存在多个抠图抠图问题，比较实验中没有局部进行比较来说明抠图算法在解决某个具体问题场景的性能，因此本文提出一套较全面的闭合形式抠图算法评价方案对选取的三种闭合形式抠图算法评价比较，首先通过理论分析算法的时间复杂度和空间复杂度，然后分类选择标准测试集中代表不同场景的图像作为评价测试集，通过实验比较抠图算法在不同场景下的抠像精度、稳定性和效率三个方面。通过评价比较得到共享抠图算法适合解决前景人像细丝毛发的问题，并且在三个算法中稳定性较好、效率较高。因此选择将共享抠图算法运用到人像植入工具中。最后本文开发了一个人像植入工具，经过测试发现本文开发的植入工具得到的植入效果要好于目前主流图像处理工具得到的效果。27419 毕业论文关键词：闭合形式抠图算法；比较；人像植入工具 Comparison of Close-form Image Matting Algorithm and Development of Portrait Implantation Tool Abstract：The current mainstream image processing tool for automatic matting function of complex foreground edge, matting portrait prospect is not accurate, such as matting filament hair. In natural image matting algorithm the closed-form image matting algorithm can get the image edge details. Therefore, this paper chooses three kinds of closed-form matting algorithm: Sharing Matting, Improving Image Matting using Comprehensive Sampling Sets, KNN matting to study. The matting algorithm comparison has the following problems: (1) Comparison experiment did not compare different algorithm systematically and comprehensively from different conditions. The present study matting algorithm there are many difficult problems, such as the foreground color and background color are similar, the foreground edge is complex, so it is important to evaluate the matting algorithm in different scenes in closed-form matting algorithm comparison. (2) There might be several matting problem in one test image, thus local comparison can accurately illustrate the performance of the algorithm in solving a specific matting problem, there is no local compare the comparison experiments or no local quantitative comparison method of matting algorithm comparison. This paper presents a comprehensive evaluation scheme of closed form matting algorithm and selects three kinds of closed-form matting algorithm to evaluate. Firstly, this paper analysis of the algorithm's time complexity and space complexity in theory. And Standard test images representing different scenes and classified as the evaluation test set, this paper compare the matting algorithm in different scenarios of accuracy, stability and efficiency of the three aspects. The result is that shared matting algorithm is suitable to solve the problem of hair filaments prospects, and it has good stability and high efficiency among three algorithm. So this paper choose sharing matting algorithm to apply in the portrait implantation tool. Finally this paper develops a Portrait Implantation Tool, and it tests that the result is better than the current image processing tools. Key words: Close-form Image Matting; Comparison; Portrait Implantation Tool 目 录 摘要 １ 关键词 １ ABSTRACT １ KEY WORDS １ 1 绪论 ２ 1.1 研究背景 ２ (责任编辑：qin) |