Research On BOW Based Image Retrieval Scheme

Various fields such as military,medical,entertainment and life are generating a large number of images at an increasing rate Limitations of storage space and computing power have led many users using local devices to outsource image processing to cloud servers.The protection of private data has become an important issue as the cloud servers are usually not completely available trust.JPEG format images are the most used format in daily life.This paper proposes an encrypted JPEG image retrieval scheme based on BOW(bag-of-words).This scheme extracts effective image features from encrypted JPEG images to reduce the workload of image users and achieves secure retrieval.The main research contents of this article are given at the next part:(1)An encrypted JPEG image retrieval scheme based on local features is proposed.In this solution,the content owner first obtains the DC and AC coefficients of the color component of the image by the discrete cosine transform.Then with the steps of stream encryption,scrambling encryption,and value replacement,the JPEG images are encrypted.At last,the encrypted files are transmitted to the server.The server can extract local statistics about the AC coefficients,use the statistics in conjunction with the BOW model to calculate the similarity,and finally return the closest image to the authorized user with no knowledge of the image.The experimental results show that the scheme in this paper can guarantee better time efficiency,security and accuracy,and support image retrieval while ensuring image security.(2)An encrypted JPEG image retrieval scheme based on QDCT coding is proposed.JPEG images are obtained through a series of steps such as DCT(Discrete Cosine Transform)transformation,quantization,and entropy coding.This paper first converts the image from the spatial domain to the quaternion domain.By analyzing the algebraic relationship between QDCT and DCT,a QDCT quantization table and QDTC coding for color images are proposed.The compressed image data is encrypted through block replacement,intra-block replacement,single table replacement and stream cipher steps.Finally,the similarity is measured by the Manhattan distance between the original image and the query image,and the distance is calculated by the two feature vectors generated by BOW on the cloud server side.The results show that the safety factor and retrieval accuracy are further increased.Through security analysis and experimental results,this paper proves that the ciphertext JPEG image retrieval scheme based on QDCT encoding is more secure than the previous scheme based on ul ? feature extraction for JPEG images and achieves better performance.