Research On Geometrical Invariant Copy Detection Based On Image Features

With the development of high-performance computing and popularity of theInternet, digital content (video, images, etc.) will be easily tampered and distributedby a variety of ways. Therefore, in order to protect the registered digital image fromillegal use, image copy detection is very important for copyright protection. Its keyissue is to extract robust feature to detect images with variety of attacks. There arelarge numbers of copy detection feature extraction algorithms, they are robust forgeneral signal attack, but are lack of robust for geometric attacks, such as rotation,cropping, scaling and translation. So to extract a geometrical invariant feature is adifficult issue that needs to be addressed.In this paper, a robust algorithm based on a similar-sector model and DiscreteFourier Transform (SSEC-DFT) is proposed. An image is first divided by48concentric ellipses and48similar-sectors as the similar-sector model does, then1DDFT magnitude spectrum, which is obtained from the pixel gray values in eachelliptical ring, is used as our image feature for copy detection. At last, the correlationcoefficient is calculated to measure the similarity between the test image and theoriginal image. The experimental results demonstrate that the detection efficiency andthe detection accuracy of the SSec-DFT scheme are improved compared to theexisting classical algorithm. And the SSec-DFT approach performs particularly wellon detecting geometric distorted images, for instance, it can detect the image which isscaled, cut or rotated to any degree.An algorithm based on the local Hu moments (Cir-Hu) proposed. The local areais from the normalized grayscale image as the circular template, and then the H umoments descriptor is identified as the characteristics of the image which is extractedfrom every local area. Finally, the measure of the distance between images is given.The algorithm has a high recall and a high precision, which is robust to commonsignal processing and geometric attacks, especially to cropping image which containslittle region of their original images and circulation translation.