Affiliation:
1. National Taiwan University, Taiwan
2. University of California, Berkeley, Berkeley, CA
Abstract
In this article, we propose a novel personalized ranking system for amateur photographs. The proposed framework treats the photograph assessment as a ranking problem and we introduce the idea of
personalized ranking
, which ranks photographs considering both their aesthetic qualities and personal preferences. Photographs are described using three types of features:
photo composition
,
color and intensity distribution
, and
personalized features
. An aesthetic prediction model is learned from labeled photographs by using the proposed image features and RBF-ListNet learning algorithm. The experimental results show that the proposed framework outperforms in the ranking performance: a Kendall's tau value of 0.432 is significantly higher than those obtained by the features proposed in one of the state-of-the-art approaches (0.365) and by learning based on support vector regression (0.384). To realize personalization in ranking, three approaches are proposed: the feature-based approach allows users to select photographs with specific rules, the example-based approach takes the positive feedback from users to rerank the photograph, and the list-based approach takes both positive and negative feedback from users into consideration. User studies indicate that all three approaches are effective in both aesthetic and personalized ranking.
Publisher
Association for Computing Machinery (ACM)
Subject
Computer Networks and Communications,Hardware and Architecture
Reference48 articles.
1. A framework for photo-quality assessment and enhancement based on visual aesthetics
2. A Computational Approach to Edge Detection
3. Learning to rank
4. LIBSVM
5. Two-dimensional discrete Gaussian markov random field models for image processing;Chellappa Rama;J. Institut. Electron. Telecomm. Engin.,1989
Cited by
13 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献