图像描述中全局特征的应用研究
首发时间:2018-12-18
摘要:图像描述是当前人工智能研究中的热门问题,它将计算机视觉与自然语言生成联系起来,目标是实现自动生成符合图像内容的描述文本。图像描述中常常通过迁移图像识别等领域中的深度卷积网络实现图像的特征映射。本文对常用卷积网络提取得到的图像全局特征进行研究,将不同卷积网络、不同深度卷积层特征应用到图像描述网络,并就特征融合方案进行探索。结果表明,通过特征融合可以有效提升图像描述效果。
For information in English, please click here
Research on Global Feature Application in Image Captioning
Abstract:Image captioning is a hot issue in current artificial intelligence research. It links computer vision with natural language generation to automatically generate description text that conforms to image content. Image caption network is often embedded with deep convolutional networks migrated from tasks like image recognition.In this paper, the global features of images extracted by common convolutional networks are studied. Different convolutional networks and convolutional features from layers of different depth are applied to the image caption network, and the feature fusion scheme is explored. The results show that the feature fusion can effectively improve the image description results.
Keywords: pattern recognition and intelligent system image captioning feature representation
基金:
引用
No.****
同行评议
勘误表
图像描述中全局特征的应用研究
评论
全部评论0/1000