Image Style Classification based on Learnt Deep Correlation Features

• Main Authors: Wu, Yi-Ling, 吳依玲
• Other Authors: Wei-Ta Chu
• Format: Others
• Language: en_US
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/n4v2n3

碩士 === 國立中正大學 === 資訊工程研究所 === 105 === Despite various studies on visual features and semantic concept detection, some image properties are difficult to extract, for the purposes of image/video classification or retrieval. Some bio-inspired properties, like sentiment and emotion, are apparently perceived by human, but are hard to be modeled in a computational way. In this work, we focus on image style property that emerges recently and is believed to be a promising extension of current classification/retrieval works. This paper presents a comprehensive study of deep correlation features on image style classification. Inspired by that correlation between feature maps can effectively describe image texture, we design and transform various such correlations into style vectors, and investigate classification performance brought by different variants. In addition to intra-layer correlation, we also propose inter-layer correlation and verify its benefit. After showing the effectiveness of deep correlation features, we further propose a learning framework to automatically learn correlations between feature maps. Through extensive experiments on image style classification and artist classification, we demonstrate that the proposed learnt deep correlation features outperform conventional CNN features and handcrafted deep correlation features by a large margin, and achieve the state-of-the-art performance.