面向海量图像高速拷贝检测的视觉指纹提取与匹配

项目名称： 面向海量图像高速拷贝检测的视觉指纹提取与匹配

项目编号： No.61003163

项目类型： 青年科学基金项目

立项/批准年度： 2011

项目学科： 能源与动力工程

项目作者： 高科

作者单位： 中国科学院计算技术研究所

项目金额： 7万元

中文摘要： 图像拷贝检测技术是实现多媒体内容监管和版权保护的关键技术。如何通过高速提取和匹配具有高稳定性和区分性的图像视觉特征，实现复杂形变下的海量图像高速拷贝检测，是亟待解决的技术难点和当前的研究热点。我们针对这些问题进行了深入研究，并获得了一系列研究成果。首先，为了提高视觉特征对各种图像变形尤其是仿射形变的鲁棒性，我们提出了一种具有高稳定性和区分性的图像特征- - 视觉指纹的提取方法。该方法通过特征选择和优化，在降低内存消耗的同时，显著提高了严重仿射形变下图象拷贝检测的准确度。在此基础上，我们基于GPU 架构和算法并行化实现了视觉指纹的高速提取。相关研究成果发表在IEEE Transaction on Multimedia等多篇该领域顶级国际期刊和会议中，并已在实际系统中获得成功推广和应用。

中文关键词： 图像拷贝检测；视觉指纹；局部特征；GPU

英文摘要： Content-based Image copy detection is one of the key techniques of multimedia content monitoring and copyright protection. How to extract robust and distinctive image visual featrues effectively,thus to realize real-time image copy detection for huge amounts of images under strong image transformations, is one of most the difficult and hot spots in multimedia content analysis and information retrieval.In order to improve the robustness of visual features under various image transformations, especially affine transformations casused by viewpoints change, we propose a novel kind of image feature called visual fingerprint which is both much more robust and distinctive than exsiting methods. Through feature selection and optimization,our method greatly improve the detection performance of image copy detection even under serious affine distortions,without sacrificing memory cost. Moreover, we also study high-speed feature extraction method through parallel processing in GPU architecture. These research results have been published in many famous international journals and conferences, such as IEEE Transaction on Multimedia,and have been sucessfully applied in many practical systems.

英文关键词： image copy detection;visual fingerprint; local features; GPU