Research On Image Retrieval Based On Region Of Interest And Bag Of Feature

With the fast development of the Internet and the popularity of various camera devices,as an effective information carrier,image has been appeared in different fields.The rapid growth in the amount of digital images has highlighted the importance of effective retrieval approaches in order to facilitate the searching and browsing of large databases.The region based image retrieval is shown to be more efficient in reflecting the user requirement in the literature.Therefore,region of interest(ROI)extraction is an important step in deriving visual features.On the other hand,Bag of Feature(BOF)model is a widely used feature in image retrieval and classification.However,BOF feature is a global feature and cannot be used for region based image retrieval directly.In this paper,we propose a general framework for ROI based image retrieval.The main works are summarized as follow:(1)We propose a new ROI extraction method based on Simple Linear Iterative Clustering(SLIC)superpixel segmentation and the saliency model.And center bias has been used for optimizing the extracted ROI.Finally,we extract a feature vector including HSV color histogram and joint statistical to represent each image.We compare the proposed method with several existed image retrieval methods on four public image databases,Corel 1K,Corel 5K,OT-scene and MIT Vistex.The experimental results show our method of ROI extraction can quickly extract the ROI and the image retrieval performance can be improved by using the extracted ROI.(2)We give a new image representation method based on ROI and image block regions.We divide an image into four blocks of fixed size and ROI.The BOF histograms of ROI and image block regions are computed by BOF model,the obtained BBOF histogram is more stable and recognizable.BBOF histogram and the low level features of the optimized ROI are used for image retrieval on four datasets.The experimental results reveal the proposed method can improve the retrieval performance to a certain degree.