Research On Semi-fragile Watermarking Technology Based On JPEG2000

As the quick development of internet and multimedia, digital information can spread rapidly and conveniently on internet, through different forms. Multimedia technology is becoming an important way of exchanging information day by day. People can communicate with each other and do business through internet. So the security of information is becoming more and more important. With coming of information society, digital information and network have been part of our life. They bring convenience and also the trouble, which refers to information can be stolen, distorted, illegal copied and spread easily.Watermarking technology is quickly growing in this situation. It's a technology that embeds marks into the digital content. The embedded marks are usually invisible or insensible, but can be detected by calculations. The watermark is combined with the data content and is inseparable part of it. Some of the watermark is robust to the accidental changes during the data transmission and publication, such as pattern transformation, compression and so on. The watermark can be made to stand for the owner, date or limits of authority, just like the brand of products. So different purpose make the information contained in watermark different.Image-watermarking is an important area among current watermarking fields. For image-watermarking, there are always many redundancies, and this is the presupposition of watermarking system, but in the view of image compression, redundancies of images are the parts to cut off. So the combination of compression and watermarking is becoming a more and more important research area of watermarking.JPEG2000 is a new set of compression standard. It integrates four models of JPEG into one. On the side of encoding, it compresses images with the largest compression quality and largest resolution. On the side of decoding, it can generate images by decoding signals in any quality and resolution. There are many advantages of JPEG2000, high compression with low data, supporting both no-loss compression and loss compression, transforming data gradually, encoding on the interested areas, tolerance of errors, open frames and so on.Because of the advantages of JPEG2000, image-watermarking based on JPEG2000 becomes a hot spot. This paper is discussed on this background, and finally designs a semi-fragile watermarking system base on JPEG2000.Semi-fragile watermarking is an important branch of watermarking. It can resist certain operations which is not ill-intention such as compression. At the same time, it's fragile with the distortion and can locate the areas that are distorted, even recover it.In this paper, the semi-fragile watermarking is robust for JPEG2000 compression, and in order to achieve this goal, we need to combine JPEG2000 compression and watermarking systems efficiently. JPEG2000 compression has a very good performance, which is a strong attack for the watermark. Specifically, the compression distortion occurs in quantization and the process of coding stream organizations. Obviously, the watermark should be embedded in the code after the quantization or after the encoding. If the watermark is embedded after encoding process, the watermark can be completely saved after JPEG2000 compression, but because of the complexity of the coding system, it would be very difficult to extract the watermark in the decoder. So we generate and embed watermark information before encoding and after quantization. At this time, watermark information is fully robust to quantization. And also the distortion during the coding stream organization is the result of discarding the blocks which are not important to the contribution of the image quality. So the watermark can be resistant to JPEG2000 compression attacks if it is embedded in larger amplitude of the wavelet coefficients.During the watermark generation, because the low-frequency wavelet transform coefficients stand for the basic content of the image, and can be completely reserved after JPEG2000 compression, so the watermark information in this article is generated from this part. We do three times of wavelet transform to the image, and input the LL₃ coefficients and the sub-block indexes as parameters for hash computing, so we get different watermark information for each sub-block. Watermark information generated at this time is too long. It will reduce the non-visibility if we embed all of the watermark bits, so finally we choose the last three bits as watermark information for embedding.At the time to embed watermark, the most important issue is the choice of coefficients for watermark embedding. According to the system of this article, coefficients to embed watermark should not change after the JPEG2000 compression, but are fragile for other attacks. Based on the statistical analysis on sub-band coefficients after three-level wavelet transform, most of coefficients in HL3, LH3 and HH₃ areas have a large amplitude, so coefficients of this part are suitable for watermark embedding. As a result, we select the part to embed watermark information. First of all, we sort the coefficients which have larger amplitudes after three-level wavelet transform in descending order, and save the relationship between coefficients and indexes. Then we embed the watermark information of the first image sub-block into the largest amplitude coefficient, and watermark of the second image sub-block is embedded in the next coefficient, and so on. The specific way to embed watermark is to modify the last three bits of the embedding coefficient. These coefficients can be preserved even after JPEG2000 compression, so the whole watermarking system is robust to JPEG2000 compression. At the same time, watermark information is abstracted from single image sub-blocks, so if we embed the authentication watermark into the image, the sub-blocks can detect itself according to the mapping relationship.The process of watermark extraction is the anti-process of watermark embedding. Decoding the compressed bit stream, we get the image data embedded with the watermark. The coefficients embedded with watermark can be found using the mapping relationship, and then the watermark information for each sub-block can be abstracted from the lowest three bits of coefficients. At the same time, do hash on low-frequency coefficients and the image sub-block index, so it can generate authentication bits from the image again. Comparing the two parts of watermark information, if they are consistent, we consider the image has not been maliciously tampered; if not, the image has been tampered with. Since the watermark is produced and embedded based on a separate sub-block of the image, it is easy to position the image block which has been tampered with.The experiments show the system is practical and reliable. It's robust to certain compression rate of JPEG2000 and fragile to malicious attacks.