Research On Image-aware Hashing Technology Based On Region Of Interest

Text and images have always been the two main ways for people to convey information.With the advent of information technology,people prefer to use images instead of words to express information faster than textual narratives.At the same time,the development of computer software,for example,Photoshop and other image editing tools have emerged to make image modification easier.Secondly,images are affected by channel noise in the transmission process.At the same time,in the daily image transmission process,often the transmission is not the original image,but the edited image.For example,the brightness of the image is adjusted,denoised and corrected,etc.Image-aware hashing technology uses hash vectors to represent images and finds the corresponding images by matching the hash vectors,which can be better applied to real life.At present,image perception hashing techniques have made some progress in distinguishing visually identical and different,but there are still problems such as no distinction between image targets,which cannot well match the human intuitive perception of images.In this paper,according to the mechanism of human visual perception-humans generally focus only on the region of interest(ROI,Region Of Interest)of an image.Meanwhile,most of the existing image perception hashing algorithms only consider the perceptual hash vector extraction on 2D images,and there is no extraction scheme for the depth image based rendering(DIBR)3D images.As a method for extracting regions of interest,the saliency map model in human eye visual feature-based detection schemes can represent the target information in images well.The existing perceptual hashing algorithm based on image saliency map does not solve the problem of not being robust to image rotation operations.Based on the above viewpoint,the image perceptual hashing algorithm based on image saliency region is proposed.The main research work as well as the innovation points of this paper are as follows.(1)A visual saliency map algorithm(Abosorbing Markov chain,AMC)based on the absorption Markov chain is proposed as a template for improvement to enhance the robustness to rotation operations.(2)Propose an improved AMC model based on the improved AMC with NMF image perception hashing algorithm.The improved AMC model basically achieves robustness to image rotation.The pixel data in the region of interest are arranged in order and a quadratic image is constructed,and NMF operations are performed on the quadratic image and the original image respectively to obtain two coefficient matrices,and the two are concatenated to form the final hash vector.While achieving robustness to image rotation,the image tampering detection operation is also achieved,and at the same time has certain advantages in comprehensive performance compared with existing image-aware hashing algorithms.(3)An image-aware hashing algorithm based on the combination of the improved AMC model and feature points is proposed.This approach filters the feature points in the outer rectangle of the region of interest and uses the descriptor and key inner product to construct the final hash vector.While achieving robustness to image rotation,the image tampering detection operation is also achieved,while having certain advantages over existing image-aware hashing algorithms in terms of comprehensive performance.(4)An image-aware hashing scheme based on the improved AMC model and pixel grouping is proposed for application to DIBR 3D images.The scheme extracts the region of interest and the whole image pixel separately,constructs the grayscale histogram separately,and then performs the grouping operation.Finally,NMF decomposition is applied to the secondary image constructed by grouping to obtain the final hash vector.The experimental results show that the algorithm proposed in this chapter has good robustness and distinguishability on DIBR 3D images with some tamper detection capability.It has better performance compared with existing perceptual hashing algorithms that rely on invariant central viewpoints.