Abstract: To bridge the semantic gap between low-level visual feature and high-level semantic concepts has been the subject of intensive investigation on big data management for years, in order to improve the accuracy of image auto-annotation. Multi-features fusion for image auto-annotation based on DBN model is proposed, combining image visual features with different weights as inputs of DBN model and optimize parameters of DBN, achieving image automatic annotation on big image data. The experimental results based on Corel image database show that the proposed method, taking different features of images into considering, has good performance on annotation precision.

Key words: multi-features fusion, deep learning, restricted Boltzmann machine, image annotation

摘要： 缩小图像低层视觉特征与高层语义之间的鸿沟，以提高图像语义自动标注的精度，是研究大规模图像数据管理的关键。提出一种融合多特征的深度学习图像自动标注方法，将图像视觉特征以不同权重组合成词包，根据输入输出变量优化深度信念网络，完成大规模图像数据语义自动标注。在通用Corel图像数据集上的实验表明，融合多特征的深度学习图像自动标注方法，考虑图像不同特征的影响，提高了图像自动标注的精度。

关键词: 多特征融合, 深度学习, 受限玻尔兹曼机, 图像标注