基于深度学习的图像检索系统

基于内容的图像检索系统关键的技术是有效图像特征的获取和相似度匹配策略.在过去，基于内容的图像检索系统主要使用低级的可视化特征，无法得到满意的检索结果，所以尽管在基于内容的图像检索上花费了很大的努力，但是基于内容的图像检索依旧是计算机视觉领域中的一个挑战.在基于内容的图像检索系统中，存在的最大的问题是“语义鸿沟”，即机器从低级的可视化特征得到的相似性和人从高级的语义特征得到的相似性之间的不同.传统的基于内容的图像检索系统，只是在低级的可视化特征上学习图像的特征，无法有效的解决“语义鸿沟”.近些年，深度学习技术的快速发展给我们提供了希望.深度学习源于人工神经网络的研究，深度学习通过组合低级的特征形成更加抽象的高层表示属性类别或者特征，以发现数据的分布规律，这是其他算法无法实现的.受深度学习在计算机视觉、语音识别、自然语言处理、图像与视频分析、多媒体等诸多领域取得巨大成功的启发，本文将深度学习技术用于基于内容的图像检索，以解决基于内容的图像检索系统中的“语义鸿沟”问题.

Image Retrieval System Based on Deep Learning

Learning effective feature representations and similarity measures are crucial to the retrieval performance of a content-based image retrieval system. In the past, the system works on the low-level visual features of input query image, which does not give satisfactory retrieval results, so, despite extensive research efforts for decades, it remains one of the most challenging problem in computer vision field. The main problem is the well-known "semantic gap", which exists between low-level image pixels captured by machines and high-level semantic concepts perceived by human. In the past, the content-based image retrieval system only works on the low-level visual features, which cannot solve "semantic gap" issue. Recently, the fast development of deep learning brings hope for the issue. Deep learning roots from the research of artificial neural network. In order to form more abstract high-level, deep learning combines low-level features, finds the regularities of distribution, which is different from other algorithm. Inspired by recent successes of deep learning techniques for computer vision, speech recognition, natural language process, image and video analysis, multimedia, in this paper, we apply deep learning to solve the "semantic gap" issue in content-based image retrieval.