Research On Geometrical Invariant Image Copy Detection Algorithms

Nowadays, with the development of Internet and multimedia technologies, more and more multimedia contents are available through various online services. The rapid growth of the multimedia services has created the potential demand for the protection of Intellectual Property Right (shortly named IPR) since digital multimedia content is easily full-copied and distributed. Generally, there are two approaches to protect copyright on multimedia content; watermarking and content-based copy detection. Watermarking embeds information into the image prior to distribution. Thus, all copies of the marked content contain the watermark, which can be extracted to prove ownership. While content-based copy detection schemes extract signatures from the original image and test ones, then store into signature database, which are used to determine whether they copies or not. In content-based copy detection, three important factors, robustness, discrimination and efficiency should be considered. At present, the most of content-based copy detection algorithms extract global feature, which handle the common signal processing well. However, global features are sensitive to geometric distortion, especially rotation and cropping, which are the difficulties in geometric-resistant copy detection research.Most of the current existing block-based image copy detection algorithms extract image signature from blocks in pixel domain. However the blocks information changes in geometric attacks. The block-based signatures are not able to resist scaling and rotation attacks. Thus, based on the properties of full DCT, an ordinal measure of full DCT coefficients (shortly named OM-FDCT) image copy detection algorithm is proposed. The full DCT coefficients are changing regularly with scaling, then the low and middle frequency DCT coefficients are selected and their ordinal measure is computed as image signature. In comparison to the two traditional block-based schemes, the robustness is verified. The experimental results show that OM-FDCT are robust to scaling and rotation with angle less than 30° To achieve high precision and recall with respect to robustness and discrimination, a multi-resolution histogram (shortly named MRH) descriptor based image copy detection scheme is proposed. Spatial information of an image is related robustly to weighted averages of the rates of change of histogram densities, while the weighted averages are the Fisher information measures. That is, the histogram can be transformed into a vector of generalized image entropies. The rates at which the histogram bins change with image resolution can be transformed into the rates at which generalized entropies change with image resolution. The change rate of the generalized entropies with image resolution is given by the generalized Fisher information measures. In a word, the multi-resolution histogram encodes spatial image information directly. The experimental results demonstrate that our method can resist to both geometric distortions and signal-processing like attacks. P-R curves obtained under various attacks show that the performance of our method is better than such a few classical methods selected for comparison.A novel image copy detection scheme based on Scale Invariant Feature Transform (shortly named SIFT) detector and multi-resolution histogram descriptor is proposed. In this novel algorithm, robust interesting points are firstly extracted using the SIFT detector and steady ones are selected by prior attacks, then a series of robust, homogenous and large size circular patches are constructed for each interesting points, finally the MRH is introduced to generate a discriminative feature vector for each patch. Experimental results obtained from the benchmark attacks demonstrate that the performance of the proposed approach is better than existing methods, especially on the test against geometric distortions.