基于内容的图像检索论文：鞋印图像检索算法的研究及系统实现.docVIP

基于内容的图像检索论文：鞋印图像检索算法的研究及系统实现 【中文摘要】鞋印图像是案发现场一种常见的图像,在刑侦科学中,常作为证据来使用。但是,随着刑侦科学的发展,从现场收集的鞋印图像越来越多,因此,如何有效、快速的从大规模鞋印图像库中检索出所需的图像是当前具有挑战性的研究课题。本文主要围绕基于内容图像检索中图像特征提取这一关键技术展开研究,内容主要包括鞋印图像的纹理特征、形状特征。首先针对鞋印图像普遍存在残缺的特点,在提取纹理特征之前采用了图像分区及分区有效性评价算法,只有当分区为有效分区时,才提取特征并参与检索。在纹理特征提取方面,采用了两种鞋印图像纹理特征检索算法：改进的基于Log-Gabor滤波器簇的纹理特征检索算法和基于傅氏变换自配准性的投影傅氏幅度谱纹理特征检索算法；在形状特征提取方面,介绍了一种基于形状区域匹配的鞋印图像检索算法。同时,在分析和研究了大量鞋印图像的特征的基础上,介绍了一种基于形状特征和统计特征的鞋印图像分类算法,将鞋印图像分为适合于提取纹理特征的类型与适合于提取形状特征的类型,以便在特征提取时采用适合的方法。最后,基于以上研究,设计并实现了一个算法验证性的鞋印图像检索系统。该系统采用Access2003作为数据库,在VC6.0下编程实现,可以方便的利用本文中介绍的各种算法进行图像检索。 【英文摘要】Shoeprint image are often left at crime scenes and can be used as scientific evidence in forensic science. Recent developments in forensic science have resulted in large numbers of scene of crime images being collected for recording and analysis, so rapid and effective searching for desired images from large-scale shoeprint image databases becomes an important and challenging research topic.In this thesis, the exploratory research work has been done around the feature extraction, which include texture, shape. For the widespread incomplete of shoeprint image, the shoeprint image is divided into several parts before extracting the texture features and evaluate the effectiveness of each part. Only when the partition is a valid part, we extract texture features and participate in retrieval. On texture feature extraction aspects, two retrieval algorithms are presented. One is improved texture feature retrieval algorithm based on Log-Gabor filter cluster, another is projection Fourier amplitude spectrum texture feature retrieval algorithm based on auto-registration of Fourier Transform magnitude spectra. On shape feature aspects, a kind of shoeprint image retrieval algorithm based on the match of shape area is proposed.At the same time, a kind of shoeprint image classification algorithm based on shape and statistical feature is presented. In this algorithm, shoeprint image is divided into the type of suitable f