Research On Image Retrieval Based On Fusion Of Multi Features

With the rapid development of mobile internet, the number of digital images isgrowing explosively, which leads to the increase of demands for image retrieval. As theresult, content based image retrieval (CBIR) becomes an active research area toautomatically index and retrieve images based on their visual content. In recent years,CBIR become focus of research and is applied to many fields.In this dissertation, the exploratory research work has been done around thelow-level feature extraction, which includes color, shape, spatial features and so on. Themain contributions of this dissertation are summarized as follows.Several key techniques and algorithms of CBIR are deeply analyzed anddiscussed, such as, the low-level feature descriptions including color, shape, texture, andposition, the similarity measure between features and evaluation methods of imageretrieval algothrims.An image retrieval algorithm FCEH, based on fusion of color, texture, andposition features, is proposed. FCEH, abbreviated from "Fuzzy Color and EdgeHistogram" and inspired by CEDD, not only combines fuzzy color histogram descriptorby Vedran Ljubovic and edge histogram descriptor by MPEG-7standard, but also fusesposition information through dividing an image to sub-images. Experiments results haveshown that FCEH is better than other image retrieval algorithms according to ANMRR.A content based image retrieval system GIRS is implemented, using Python andOpenCV. GIRS is recursive acronym for "GIRS is an Image Retrieval System".Integrating several descriptors and providing image retrieval and algorithm evaluationinterfaces, GIRS can be used as the basic platform for research on CBIR. Besides, GIRSis a general purpose image retrieval system like Google and Baidu for its compact andfriendly user interface.