Image Hashing Algorithm And Implementation Based On Image Feature

With the rapid development of the Internet and multimedia globalization, the use of digitalimages has become extremely popular. Multimedia information communication today istherefore no longer limited by time, space, and location. Anyone, even malicious users, mayeasily obtain information which they need anytime and anywhere through open networks. Inthe process of Internet information transmission, malicious attacker can tamper information orsteal it arbitrarily. The widespread piracy images, audio, video files and other multimediainformation have suffered from illegal access and unauthorized distributions. Such actions notonly harm the legitimate rights of copyright owners but also lead to a reliability crisis forinformation networks. Most of the time, tampering and stealing is difficult to discerndistinction by naked eye alone, we must adopt necessary science technical ways to ensure itsauthenticity and credibility. At present, technical means to ensure the security of digitalmultimedia information have two main aspects. Except for digital watermarking for protectingintellectual property rights, another way is using digital hashing techniques. Different fromdigital image watermarking, where a digital image watermark signal is embedded into the hostimage, the digital image hash depends on the image content itself, requiring no embedding.The benefit is achieved at the cost of exclusively storing the signature/hash for each image inthe database. How to save and search image hashes for a large-scale image efficiently, whichhas become one of our current main task. Additionally, the robustness of the hash is also animportant aspect which we need to consider. During image transmission, it often experiencessome routine operations which does not change the image’s content, for example, JPEGcompression, channel additive noise, brightness changes, rotation and filtering, etc. Therefore,in the process of image authentication, not change significantly as long as the image contentand image visual quality, we think the image is legitimate. However, it will turn a blind eye onmalicious tampering towards the image if the hash’s robustness is too low. We should reconcile the conflict between the robustness of the hash and its security or sensitivity.However, a new image hash scheme is proposed in this study. Our scheme mainly dependson the characteristics of robust Harris corner points and Fourier polar phase spectrum. Themeans takes advantage of the merits of image airspace domain Harris corner and polar Fouriertransform phase spectrum. We extract hash from the original image in two parts separately.And then put the two parts of hash into a new hash. Robust Harris corner points of the imagecan improve the robustness of the operation, which renders the hash strong robustness to largegeometric operations. The part of hash component extracted via corner points is robust to largegeometric interference robustness ideally. But its robustness is not ideal in dealing withGaussian noise, additive noise, filtering and median filtering. When the entire image onlysuffers unmalicious geometric attacks, this part of component in the synthetic hash doesn’tchange significantly. Thus the synthetic hash is also robust to large geometric attack better.The main steps are: Firstly, we must determine robust Harris corner points which are robust tothe image rotation. Secondly, we extract the information around these feature points and usesingular value decomposition about them, generate a part of component in the synthetic hash.This part of the hash extracted by Harris corner points can reflect the content of the image to acertain extent. Singular value of corresponding position also changes if the image contentchanges. In this sense, the tampering of the image content will affect a change of image hash.Meanwhile, another part of component obtained by Fourier transform phase spectrum isrobust against JPEG compression and filtering routine operation or the shear and replacemalicious tampering operation. But its robustness is not ideal in dealing with a certain degreeof the geometric distortion. Therefore, in order to combine the advantages of the two parts, wereconstruct each hash obtained by the two parts into a new hash. We make sure that the lengthof the hash is appropriate and not very larger or much smaller. Hash obtained by this method isimproved in image security and robustness significantly. Digital image phase information hashigh sensitivity in the process of generating the digital image hash. The image hash’scomponent obtained by the robust Harris corner points can be merely robust to geometricattacks. But for the image noise and filtering, and its reaction is too excessive to make image’ssecurity be reduced. In order to improve this situation, another part of component in the imagehash are obtained by Fourier transform phase spectrum, which are robust against JPEG compression and filtering or tampering routine operation. Fourier transform phase spectrumwhich can position each frequency component of the image of accurately. In physical sense,the amplitude spectrum determines the number of the frequency components of an image. Thephase spectrum determines the position of each frequency component in the image. The imagecan be completely preserved as long as each frequency component is stored in the correctposition. The image becomes distorted beyond recognition if we change the phase informationof the image.The contributions of this study are as follows:1) the hash is robust against imagetampering and malicious attacks；2) the image signature extracted by our scheme helps toconserve storage space. Combining with the test data, we will analyze the scheme’sperformance from distinguish, security and robustness these three aspects in the experiment. Inthe experiment, we selected three different types of attack methods to analyze the safetyperformance of the proposed algorithm. In this process, our algorithm is compared with MinWu’s two algorithms and the NMF algorithm. Meanwhile, we will analysis some conventionaland malicious operations which the image is experienced in the channel during transmission,such as JPEG compression, noise, filtering and etc. Experimental results show that whencompared with the current extracting hash from single domain schemes, this scheme can berobust to JPEG lossy compression, channel additive noise, brightness changes and filtering.Meanwhile, it can also deal with a large class of attacks including large geometric transforms(eg, rotation, extrude, cut, etc.) and image tampering effectively. Moreover, the image hashingis improved in image security and robustness significantly.