ABSTRACT

Recent works in image perception and multimedia technology have paved the way for automated analysis of visual arts. Intelligent painting based on image perception, processing and identification in multimedia enterprise is the trend. In this paper, we use a Cross- Contrast Neural Network (CCNN) model to automatic art identification in oil painting. To achieve this point, we first retrained a tailored CNN network to extract features of oil paintings and then calculated the cross-contrast probability map utilising the contrast information to measure the similarity between input images. We aim to combine IBS with CNN to facilitate the training progress and reduce the difficulty of finetuning the parameters of CNN. To demonstrate the effectiveness of our approach, we introduced the Selected-Wikipaintings dataset, containing over 5000 images painted by 20 artists in various styles. The average accuracy of the classification task in 20 artists achieved 85.75%, which proved to be more accurate than the general method in artists identification task. Our extensive experimental evaluation shows that CCNN owns better classification performance in small data-set multi-classification scenes.

摘要

最近在图像感知和多媒体技术方面的工作为视觉艺术的自动分析铺平了道路。多媒体企业基于图像感知、处理和识别的智能绘画是趋势。在本文中，我们使用交叉对比神经网络 (CCNN) 模型来自动识别油画中的艺术。为了实现这一点，我们首先重新训练了一个定制的 CNN 网络来提取油画的特征，然后利用对比度信息计算交叉对比度概率图来测量输入图像之间的相似性。我们的目标是将 IBS 与 CNN 相结合，以促进训练进度并降低微调 CNN 参数的难度。为了证明我们方法的有效性，我们引入了 Selected-Wikipaintings 数据集，包含由 20 位艺术家以各种风格绘制的 5000 多幅图像。20位艺术家的分类任务平均准确率达到85.75%，在艺术家识别任务中比一般方法更准确。我们广泛的实验评估表明，CCNN 在小数据集的多分类场景中具有更好的分类性能。