Image Retrieval Method Based On User Region Of Inierest

In recent years, with continuous development of information storage technology, image hasbecome a major carrier of information, so image retrieval which based on the quantity of imageshave caused people’s more and more concern. Normally image will contain much redundantinformation such as background, this paper puts forward an idea that combine the visualperception of model algorithm which can extract the ROI automatically with CBIR. By this idea,we can extract and retrieval the picture which contains more than one interested region at onetime. First, we use the Itti visual perception model to extract the interested region on the originalimage, then take used of color and texture feature to extract ROI’s feature vector, finallycomposite similar distance with degree of ROI to complete the image database retrieval. Themain research results are as follows:(1)Extraction of ROI: Based on human visual principle, the Itti visual perception modeluse Gabor filter to establish gaussian pyramid on the image of color, brightness and directionalcharacteristics, then get the corresponding multilevel image sequence. Second usecenter-surround operate the sequence charts to get significant maps of three characteristics.Finally synthesize the significant maps to get total significant diagram, thus form a series of stayattention. In this paper, the way how to select the Itti’s scale is analyzed and discussed, In orderto flag the ROI correctly and comprehensive,we improved the original Itti code on the base ofimage retrieval features. Finally give the method of how define the degree of ROI.(2)The content-based image retrieval: first compared some common feature extractionmethod’s advantages and disadvantages where based on the CBIR.Then analysis extractionmethod and its corresponding similarity measurement method which is used by this paperemphatically. Color histogram extraction algorithm need convert the original image is to HSVcolor space to statistical interval number, then use Gabor filter to filter image describe image’stexture; Use histogram intersection distance to metric color similarity, use image filter mean and variance to metric texture similarity distance. Finally use the color measurement value and thetexture measurement value with gaussian normalization, thus obtains this new image retrievalalgorithm.(3)Combine the improved code of Itti algorithm with the image retrieval, and consider asituation that when an image contains multiple region of interest. In that case introduce thedegree of ROI into the similarity measure. In order to validate the effectiveness of the proposedalgorithm in this paper, design and develop a simple image retrieval system, and carry oncontrast test from to show that this algorithm is effective.