The Research Of Image Retrieval Algorithm Based On Visual Saliency

With the rapid development of the Internet technology and the rapid popularization of the mobile terminal devices represented by smart phones,people produce a large number of multimedia data every day,in which image data occupy a large proportion.The traditional image database technology has been unable to meet the current needs,how to storage and manage the massive image data effectively has become a hot topic in academic circles.Therefore,the Content-Based Image Retrieval technology(CBIR)has become the focus of attention,and gradually become the mainstream technology of the new generation of image database.The traditional Content-Based Image Retrieval technology mostly uses the low-level features(such as color,shape and texture)to calculate the similarity of the images,so less attention is given to the visual attention of human eyes.The so-called human visual attention refers to when people observe scene around or pictures,will automatically find certain objects as interesting or important goals.The visual attention mechanism of human eyes coincides with the goal of content-based image retrievaltechnology.Therefore,it is of great significance to improve the accuracy of retrieval by fusing the visual attention mechanism of human eyes to content-based image retrieval algorithm.This paper summarizes and analyzes the new features of content-based image retrieval technology,utilize human visual attention mechanism to help retrieval algorithms improve the ability of image analysis and understanding,and improve the accuracy of the retrieval algorithm.The focus of this paper is to utilize of visual saliency map to resolve the problem of spatial location of the low-level image features,the problem of the elimination of visual words in image background in Bag of Word model and the problem of multi feature fusion in retrieval algorithm.The main works of the thesis are listed as follows:1.Based on human visual attention mechanism,an image retrieval algorithm based on visual weighted spatial pyramid model is proposed by combining the visual saliency of the image.Firstly,the image is processed by multi-scale spatial pyramid model and the feature vectors of each block are obtained.Then,the visual weight of space pyramid is calculated based on the visual saliency of human visual characteristics.In the end,we use the visual weight to process the joint feature vector of multi-scale spatial pyramid,and the retrieval could be got based on the obtained joint feature vector.2.In view of the problem of the loss of spatial location information of the visual word and the visual vocabulary affect the retrieval accurancy,the paper proposed a new visual vocabulary that can reflect the spatial location of visual word by combining visual saliency and visual word.The new constructed visual vocabulary takes into account the spatial location of the visual word,which can reflect the local features of the object in the image,and try to get rid of the influence of the background information.A large number of simulation experiments show that the new visual vocabulary can effectively improve the accuracy of retrieval algorithm.3.Content-based Image Retrieval technologies is based on a comprehensive and in-depth understanding and analysis of the content of the images,most bag of word models use a single local feature as a visual word,so it is difficult to describe the objective information of the images.In this paper,we propose an algorithm based on Sift features and color features for bag of word model.The algorithm combines the human eye's visual attention model to analyze and understand the retrieval image,and detect the target object in the image and separate it from the surrounding scene.Then,the visual vocabulary and color histogram are calculated with the extracted object.Finally,the visual vocabulary and histogram are combined to calculate the similarity between images.Experimental results show that the algorithm is effective.