照片美感品质的客观评价研究

发布时间：2018-06-03 23:13

本文选题：照片 + 计算机美学　；参考：《云南大学》2014年硕士论文

【摘要】：近几年来,随着家用相机的普及和手机自带拍照功能的逐渐增强,并且在许多社交网站和手机应用中都加入了照片分享功能,因此让更多的人喜欢上了摄影,从而导致现今照片的数量呈爆炸式增长。因此,利用先进的计算机技术来帮助人类,在基于计算机美学的相关技术之上,对各类照片进行美感评价与分析这一研究领域越来越受到各方学者的关注。传统的对照片美感品质评价仅是依靠摄影师或者观赏者们基于自身对美学和摄影理论的理解来做出主观的评判,而计算机美学的出现与兴起正是对原有照片美感品质评价的主观性和单一性做了进一步的完善。此外,通过计算机进行照片特征值的提取及机器学习和分类的过程,模拟人脑对照片美感品质的评判,建立了美学和人工智能及模式识别之间的桥梁。因此,该项研究属于摄影学、美学和计算机科学等多个学科相互交叉与融合的创新性前沿研究课题,且具有相当重要的理论研究价值以及广阔的应用前景。本论文在图形图像处理技术、摄影理论和计算机美学的背景支撑下,有效结合照片中的各种特征,共提出了三十八种能较好表征各类照片不同特点的数字化特征。实验中所用到的照片包括人像、动物、植物、静物、建筑、风景和夜景共七个类别。在对这些照片进行美感品质客观评价时,依据机器学习理论的指导,本文使用了支持向量机(SVM)、Adaboost、线性回归分类和随机森林这四种分类器,并且在进行分类实验时使用了十交叉检验的方法。在对实验结果的分析中发现,不同特征对不同类别照片美感品质的影响存在着一定的相似性,但同时也有显著的差异,如结构特征对人像照片美感品质的影响程度相比景物类照片更高。本文的主要内容有如下几个方面：首先,对计算机美学和摄影基础理论进行简要的介绍与概述,并将照片美感品质的主观判断依据和摄影基本理论以及计算机美学三者之间的关系进行充分的分析。其次,以照片美感品质的主观判断依据和摄影基本理论为基础,分别提出了照片的布局与结构、照片的亮度和照片的暗通道等多种能够较好表征照片美感品质的数字特征,并通过实验分析不同特征在对照片美感客观评价时的影响程度。最后,依据照片的类别在所提特征中选取相应特征,对各类照片美感品质在多种分类器中进行客观分类实验及结果分析,从而对照片美感做出客观评价。实验表明,使用本文所提的三十八种数字特征对各类照片进行客观的美感品质评价是行之有效的,该项研究不仅为照片的美感品质分析与评价提供了一种全新的方法,还能帮助摄影师从自己的拍摄作品中选取出其中具有较高美感品质的这些照片,或者在照片拍摄之后马上能显示其高/低美感品质从而来辅助拍摄者决定将这张照片保存或是重拍等广泛的用途。
[Abstract]:In recent years, with the popularity of home cameras and the gradual increase in the ability to take photos with mobile phones, photo sharing has been added to many social networking sites and mobile apps, so that more people are interested in photography. This has led to an explosive increase in the number of photographs today. Therefore, using advanced computer technology to help human, on the basis of computer aesthetics related technology, the aesthetic evaluation and analysis of all kinds of photographs has been paid more and more attention by scholars from all over the world. The traditional evaluation of the aesthetic quality of photographs is based on the understanding of aesthetics and photography theory by photographers or viewers. The appearance and rise of computer aesthetics have further improved the subjectivity and singularity of the original photo aesthetic quality evaluation. In addition, the process of feature extraction, machine learning and classification by computer is used to simulate the evaluation of the aesthetic quality of photographs by human brain, and a bridge between aesthetics, artificial intelligence and pattern recognition is established. Therefore, this research belongs to the innovative frontier research subject, such as photography, aesthetics, computer science and so on, which intersects and merges each other, and has quite important theoretical research value and broad application prospect. Based on the background of graphics and image processing, photography theory and computer aesthetics, this paper presents 38 digital features which can better represent the different characteristics of various kinds of photographs. The photos used in the experiment included human figures, animals, plants, still life, architecture, scenery and night scenes. Based on the guidance of machine learning theory, four classifiers, support vector machine (SVM), linear regression classification and random forest classification, are used to evaluate the aesthetic quality of these photos. And the method of ten cross-test is used in the classification experiment. In the analysis of the experimental results, it is found that the influence of different characteristics on the aesthetic quality of different types of photos has some similarities, but there are also significant differences. For example, the influence of structural features on the aesthetic quality of portrait photos is higher than that of landscape photographs. The main contents of this paper are as follows: First of all, the basic theories of computer aesthetics and photography are briefly introduced and summarized, and the relationship between the subjective judgment of the aesthetic quality of photographs, the basic theory of photography and the computer aesthetics is fully analyzed. Secondly, based on the subjective judgment of the aesthetic quality of photographs and the basic theory of photography, this paper puts forward the digital features which can better characterize the aesthetic quality of photographs, such as the layout and structure of photographs, the brightness of photographs and the dark channels of photographs. The influence of different characteristics on the objective evaluation of the aesthetic perception of photographs is analyzed through experiments. Finally, according to the category of photos in the proposed features selected the corresponding features, the aesthetic quality of all kinds of photos in a variety of classifiers for objective classification experiments and results analysis, so as to make an objective evaluation of the aesthetic perception of photos. The experiment shows that it is effective to use the 38 digital features mentioned in this paper to evaluate the aesthetic quality of all kinds of photos. This study not only provides a new method for the analysis and evaluation of the aesthetic quality of photographs. It can also help photographers to pick out these pictures that have a higher aesthetic quality from their own photographs. Or it can show its high / low aesthetic quality immediately after the photo is taken to assist the photographer in deciding to save or remake the photo for a wide range of purposes.
【学位授予单位】：云南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TB85;TP391.41

【参考文献】