Aesthetic Image Classification Based on Multiple Kernel Learning
Aesthetic image classification aims at predicting the aesthetic quality of photos automatically, i.e. whether the photo elicits a high or low level of affection in a majority of people. To solve the problem, one challenge is to build features specific to
- PDF / 371,597 Bytes
- 8 Pages / 439.37 x 666.14 pts Page_size
- 50 Downloads / 211 Views
School of Information Technology and Management, University of International Business and Economics, Beijing 100029, People’s Republic of China [email protected], [email protected] 2 Department of Computer Science and Technology, Beijing Electronic Science and Technology Institute, Beijing 100070, People’s Republic of China [email protected] 3 Department of Automation, Beijing University of Civil Engineering and Architecture, Beijing 100044, China [email protected]
Abstract. Aesthetic image classification aims at predicting the aesthetic quality of photos automatically, i.e. whether the photo elicits a high or low level of affection in a majority of people. To solve the problem, one challenge is to build features specific to image aesthetic perceptions, and another one is to build effective learning models to bridge the “semantic gap” between the emotion related concepts and the extracted visual features. In this paper, we present an approach for aesthetic image classification based on Multiple Kernel Learning (MKL) method, which seeks for maximizing the classification performance without explicit feature selection steps. The experiments are conducted on a large diverse database built from online photo sharing website, and the results demonstrated the advantages of MKL in terms of feature selection, classification performance, and interpretation, for the aesthetic image classification task. Keywords: Aesthetic quality · Image classification · Multiple kernel learning
1
Introduction
Aesthetics is a sub discipline of philosophy and axiology dealing with the nature of beauty, art, and taste. The assessment or prediction of aesthetic value in images is considered to be of subjectivity and universality. The subjective feature suggests that the judgment relies on individual personal feelings, and there is no single agreement on what it exactly belongs to. In contrast, the universality indicates that certain features in photographic images are believed to please humans more than others. In conclusion, though the evaluation of beauty and other aesthetic qualities of photographs is highly subjective, still they have certain stability and generality across different people and cultures as a universal validity to classify images in terms of aesthetic quality [2]. Figure 1 shows two photos from an online website, and according to the ratings by web users, it is confirmed that the photograph (b) can inspire higher aesthetic feelings than the left one (a) for most people. © Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part II, CCIS 547, pp. 229–236, 2015. DOI: 10.1007/978-3-662-48570-5_22
230
N. Liu et al.
There could be many applications making use of an algorithm for photo quality assessment. For example, a search engine can merge a photo aesthetic factor into its ranking stage to get most relevant and better looking photos. An advertiser can make a choice referring to the most beautiful photos selected by the aesthetic quality assessment tools. Photo management solutions,
Data Loading...
