近似重复图像检测及其应用-信号与信息处理专业论文.docxVIP

ABS丁RACTABSTRACT ABS丁RACT ABSTRACT With the rapid development of multimedia technology and Internet technology， the numbe辱of digital images Oil the Internet shows explosive growth．Compared with the text，images are easy to understand，which makes images of the wide range of applications，such as medical images，news images，commercial images and SO on． Beside the convenience of information dissemination brought by digital images，we are also facing an increasingly serious problem，that is，how to find images that we need on the Internet．The key technology to solve this problem is image retrieval． Research on the traditional image retrieval is divided into text-based image retrieval and content·based image retrieval(CBIR)．Text—based image retrieval can take use of text retrieval methods，but this approach depends on the tags of images． It’S difficult for people to get consistency on the selection of tags for images． Therefore，researchers pay attentions to content-based image retrieval．CBIR takes use of visual content of images to index them．As existence of semantic gap and curse of dimensionality,content．based image retrieval has encountered a bottleneck． However,as a branch of content—based image retrieval，near—duplicate image detection gets widely attention and study because of the specificity and well definition of the problem． This thesis focuses on the choice of distance function and the effective index of large image dataset．Also，according to good application of near-duplicate image detection，we designed an algorithm to automatically add the emotional tags to images based on it。The main content of this dissertation and comributions are as follow： 1．Summarizes the current research content and main results of image retrieval， describes the development trend of image retrieval，and introducts the near—duplicate image detection in details． 2．Presents a new distance function，called manhattan correlation distance(MR distance)，which could be applied to near-duplicate image dete