Research On High-Capacity Reversible Watermarking For Digital Images

The cover image will experience some imperceptible,irreversible or permanent distortions due to digital watermarking and the watermarked image can not be inverted back. Although the distortion is often quite small,it may be intolerable for some applications, such as in the fields of law enforcement,medical and military image systems.It is desired to reverse the watermarked image back to the original image after the hidden data is correctly extracted.In such cases,watermarking technique satisfying this reversibility requirement called reversible,invertible or erasable watermarking technique is proposed. As researches and applications go further,the traditional algorithms are severely challenged in watermark capacity and the capability of resisting attacks.The thesis mainly focuses on the key points in high-capacity reversible watermarking algorithms and reversible watermarking algorithms resistant to attacks.The main contributions are as follows:1.A reversible watermarking algorithm based on difference expansion and companding technique is proposed.The proposed algorithm changes statistics characteristic of the location map used to identify the expanded positions by a combination of difference expansion and companding technique.As a result,the compressibility of the location map is improved,and accordingly,the high embedding capacity can be achieved.By reasonably setting the threshold in the companding technique,the algorithm can maintain high PSNR values.2.High-capacity reversible watermarking by expanding the prediction-error to carry watermark information is proposed.First,the prediction-error is compressed into a value which is less than or equal to the original prediction-error,and then the resulting value is expanded to carry 1-bit watermark.The algorithm obtains a significant improvement in both embedding capacity and embedding distortions.A novel lossless data hiding algorithm based on a combination of predictor and the prediction-error adjustment(PEA) is also presented.PEA is introduced to make large prediction-error available for embedding while causing low embedding distortions,and accordingly,the location map can be compressed well.As a result,the hiding capacity is largely increased.3.A reversible watermarking algorithm based on invariability and adjustment on pixel pairs is proposed.In this algorithm,a smaller value,i.e.,half the difference plus 1-bit watermark,is added to one pixel and subtracted from the other,which will result in the invariability of the sum of the pixel pairs.Since the modification to each pixel is smaller,higher PSNR value can be obtained comparing with the traditional reversible methods.Based on the invariability of sum values and the equality between the parities of sum values and difference values,the extraction of watermarks and the recovery of pixel pairs can be easily achieved.In addition,Pairwise Adjustment Operation(PDA) is proposed to significantly reduce the capacity consumed by overhead information.As a result,the embedding capacity can be considerably increased while maintaining high PSNR comparing with the traditional reversible algorithms.4.A twice-embedding reversible watermarking algorithm is presented.In the first embedding process,without considering the potential problem on overflow/underflow in pixel values,payload is embedded into the wavelet coefficients of three high-frequency subbands,i.e.,HL₁,LH₁,HH₁.After that,those watermarked pixels really suffering from overflow/underflow are forced to locate in the permitted range,e.g.,[0,255]for 8-bit grayscale images,by histogram modification and their locations are associated with '1' in a location map.Since these locations make up a very small percentage of the location map,strong lossless compression for this map is allowed.The compressed map is embedded into the once watermarked image by improved difference expansion in the second embedding process.Especially for the natural images with pixels distributed across the entire gray-level spectrum,our algorithm can achieve high embedding capacity with high PSNR values.5.Considering that few existing reversible watermarking algorithms can resist to cropping attacks,a block-based reversible watermarking scheme resisting to cropping attacks is presented.In the scheme,a locating pattern is embedded into the carrier image by a modified patchwork algorithm.As a result,the cropped position can be correctly detected on the decoding side.The locating pattern just makes modifications to a small part of the pixels in each image block.The rest of the pixels are modified by difference expansion to carry the hash function of each original image block so that the local authentication is achieved.Comparing with the traditional modulo addition,since those to-be-flipped pixels are kept unaltered,the pepper-and-salt noise is efficiently avoided. Since image ID and block index serve as parts of input to hash function,the algorithm also can efficiently resist to collage attack.