基于Gauss滤波的多媒体图像的可视化检索

基于图像的可视化特征的图像检索和查询,是当前多媒体系统的重要研究方向。对于基于图像局部匹配的检索,传统的模板匹配和区域分割方法只能在目标图像中检索非常相似的对象,对尺度或形状有较大变化的目标则难以得到满意的结果。提出了一种不依赖于模板匹配的方法——图像的多尺度高斯滤波表示,用于图像的特征提取和局部匹配,并在分布并行索引结构下实现了该方法的快速索引和搜索算法。实验表明,该方法能够在较大尺度和视角变化范围内,更有效地进行图像的特征提取和检索,对基于图像局部区域相似性的匹配能达到较好的效果。

Image Retrieval in Multimedia Database Based on Gauss-Filtering

With the development of Internet and multimedia technology, images retrieval based on their visual features become an important research area. In this paper, we propose a new method for local similarity retrieval of images (multiscale Gauss-filtering representation, MSGF) which is more efficient than conventional template matching. Images are first filtered in several scales with Gaussian derivative filters, and the feature vectors are constructed on filtering output in several scales. Next, images are matched in a point to point style and with a stricter spatial checking. We develope a parallel distributed index structure and a fast search algorithm in order to use this method. Experiment shows that the method is effective in feature extraction and images retrieval by their local similarity.