Research On Uyghur Printed Complex Document Image Retrieval Based On Local Feature

The growing trend of digital era,the scale of electronic documents is rapidly growing.Therefore,the effective management of document data,and quickly search the desired target document image gradually become an urgent problem to be solved.Many researchers in the field of document image retrieval at home and abroad are borrowed from the effective algorithm in IR,they were combined with the text image and the nature of the layout itself to achieve itsrapid and accurate retrieval.At present,many kinds of local features with robustness have been applied in the image retrieval research.In this paper from the inspired,it was conducted retrieval research on the complex layout of the Uyghur document image,let the minority text image processing introduced to the new direction.The mixed complex layout document image is different from the natural scene,plain text image,it is not only is a kind of multimedia material that contain of text,pictures,tables and mathematical expressions with rich layout information,also the theme of large-scale text image library.In this paper,the complexlayout Uygur printed document image was taken as the object of study,it is get high-precision retrieval without the complicated layout segmentation and analysis,and it was made up for the blank in the study of Uyghur complex text image retrieval The specific contributions of this paper are as follows:(1)An irregular complex layout Uyghur text image database was constructed.(2)The SURF feature of Uighur printed complex document image was extracted and improved,which shortens the time of feature extreation and reduces the computational complexity.(3)The FLANN bidirectional matching and KD-Tree+BBF matching were implemented for SURF and modified SURF features.Under the size,rotation and brightness transformations,the matching performance of the two matching algorithm were compared and the robustness of local feature was verified.(4)The four kinds of distance based similarity measuring and the matching number based similarity measuring retrieval mathods were carried for local features.And conducted some retrieval experiments on classification,cutting and compression Uyghur complex document image database,respectively,to compared analyzed the performance of two retrieval algorithms.(5)In order to verify the stability and reliability of the two kinds of retrieval systems,Uyghur,Arabic,Chinese,Tibetan and naturel image database are constructed,and conducted some comparative retrieval experiments under various of transformation to the query image,respevtively.It can be seen that,the retrieval seystem based on the improved SURF feature can satisfies the needs of the user to find the target image quicly and accurately.