深度学习与一致性表示空间学习的跨媒体检索

提出一种基于深度学习与一致性表示空间学习的方法,针对图像与文本2种模态,分别采用卷积神经网络模型和潜在狄利克雷分布算法学习图像的深度特征和文档的主题概率分布;通过一个概率模型将两个高度异构的向量空间非线性映射到一个一致性表示空间;采用中心相关性算法计算不同模态信息在此空间的距离.在Wikipedia Dataset上的实验结果表明:在单模态输入检索中,文中方法的平均准确率为38.43%,相比于其他方法有明显提高.

Cross-Modal Multimedia Retrieval Based Deep Learning and Shared Representation Space Learning

A new learning method based deep learning and shared representation space learning is proposed in this paper. Using image and text as an example, we learn the deep learning features of images by convolution neural networks, and learn the text topic probability distribution by a latent Dirichlet allocation model respectively. Then nonlinear mapping the two features spaces into a shared presentation space by a probability model. At last, we adopt centered correlation to measure the distance between them. The experimental results in the Wikipedia Dataset show that our approach is better than that of the similar experiments for single mode input retrieval in recent years and its mean average precision reaches 38.43%.