Study Of Inverse Halftoning And Hardcopy Watermarking Schemes

Halftone images contain only two tones and they are generated from multi-tone images by halftoning techniques which are widely used in the printing process. Inverse halftoning addresses the problem of recovering a con-tone image from a halftone. With the development of digital library construction and copyright protection of printouts, inverse halftoning (IH) and hardcopy watermarking (HW) become active fields of image processing for halftone images. However, algorithms of IH available still can not fulfill the application to fast achieve high-quality reconstructed images because of the ill-posed essence of IH problem, and those of HW also can not meet the demand of high security and low cost, since halftones are special and print-scan (PS) operation is complicated.In this dissertation, to further improve the performance of IH and HW for error-diffused halftones, new schemes are proposed in vector IH, grayscale IH, HW for halftones and HW for con-tone images respectively. The utility and aptitude of IH is enhanced, the invisibility and robust of HW is improved.Analysis of vector halftones shows that it is important to remove the correlation between different color opponents, so an edge detection method for vector halftones and a vector IH algorithm based on such scheme are presented. By constructing the grayscale image that comprises all edges and designing multi-scale edge detector mask, the former method achieves the object of fast detecting edge information and refining halftone noise simultaneously. The latter uses the matrix gain model and proves that vector IH is equivalent to removing correlated high-frequency noise. Hence, correlation removing operator and denoising operator are designed respectively to gain IH image. Besides, two objective indexes of color similarity and mottle histogram are defined to evaluate color IH images.Grayscale IH algorithms are studied. To overcome Bayesian algorithm's shortcomings of requiring the knowledge of halftone kernel, high computational complexity and memory buffer, an improved method via maximum a posteriori is proposed. It designs noise preprocessor based on linear gain model and image classification scheme which can preserve the edges while smoothing the noise. To reduce the implementation cost, matrix-based operation is employed to estimate IH image, which makes the algorithm more practicable. To efficiently remove the limit cycles in dark area, an IH algorithm based on human vision system (HVS) is proposed. After constructing the quantitative relation between uniform grayscales and their halftones, noise is efficiently refined with the idea of dividing and conquering for different areas. It has the advantages of easy operation, high speed and good reconstructed visual quality. We also give an adaptive IH algorithm based on laplacian pyramid, which is good in generality. After introducing noise granularity to reflect the influence of different halftone kernels and establishing the model of transferring halftone noise to impulse noise, median filter is used to achieve IH image.This dissertation also systematicaly studies HW for halftones. For the condition of con-tone images not available in application, a direct-embedding watermarking algorithm based on HVS is presented. It takes HVS's characteristics into account and embeds watermark based on r-bit filter. R-bit filter is able to increase the watermark capacity, which may strengthen the robustness of PS attack. When multi-tone image is known, a grayscale HW algorithm based on threshold-altered error diffusion and an adaptive watermarking algorithm for color halftone in spatial domain are designed. The former indicates that halftones should have green-noise characteristic to resist PS attack, and watermark is embedded by constructing watermark screens together with altered-thresholds. The latter first constructs two sets whose color similarity is low, and then adaptively searches the optimum halftone vector in two steps according to color similarity. Experiments show the two watermarked images have uniform texture, and also their imperceptibility and robustness are enhanced to some extent. They can be used as a substitute to the printer's original halftone algorithm.Finally, a watermarking scheme for ordinary users is proposed, which enbeds watermark into con-tone images without changing halftone process. Since PS operation has little influence on frequency coefficients'polarity, by defining polarity sequence and watermark-represented binary sequence, as well as constructing polarity modulation model, watermark is embedded by modulating correlation coefficients between sequences. This method is good in watermark imperceptibility. It is resistant to light and contrast variety, cropping, image mosaic, PS attack and post-processing operations.