Research On Bag Of Encrypted Word Model Based Image Retrieval Scheme

With the development of digital imaging equipment,the number of images has increased sharply.Content-Based Image Retrieval(CBIR)is used to solve the problem of huge image database retrieval.The CBIR scheme has a huge demand for computing and storage costs,making it a hot issue to outsource CBIR methods to cloud server.However,the cloud server is not completely trusted.The existing encrypted image retrieval scheme has large workload for image owners and users.A secure image retrieval scheme based on Bag-Of-Encrypted-Word is proposed to reduce the user's computing burden.The specific researches are as follows.Research on Bag-of-Encrypted-Word model.In text information retrieval,Bag-Of-Word(BOW)model is widely used because of ignoring word meaning and word order.With the development of image retrieval technology,the Bag-Of-Visual-Word(BOVW)is applied to local features of images.Aiming at the research of secure CBIR scheme,we propose a Bag-of-encrypted-word model(BOEW),which is more suitable for image retrieval in ciphertext domain.Moreover,the model is applied to the specific encrypted image retrieval scheme.Research on color feature BOEW based scheme.This scheme protects image content by replacing image color values,block scrambling and pixel scrambling in image blocks.The local histogram is extracted from the image block,and all the locally encrypted histograms are clustered with K-Means algorithm,and the encrypted vocabulary is built.Each image is represented as the word frequency characteristic vector based on the encrypted lexicon.The combination of feature vectors and image identities is used to establish linear index,and the similarity between images is measured by distance between the features.Research on texture feature BOEW based scheme.Considering the limitations of color features.we study texture feature based scheme.This scheme protects the image content by the image difference matrix computation,difference value replacement,block difference matrix scrambling and the block difference matrix scrambling.BOEW model is built to represent the image into the feature vector based on this model.At the cloud server side,we calculate the Hamming distance between feature vectors to measure the similarity between images.This scheme establishes tree index and linear index,to compare retrieval efficiency.Through security analysis,it is proved that the BOEW-CBIR scheme is secure.The experimental results show that the CBIR scheme based on BOEW model can achieve better retrieval results.