Reversible Watermarking Based on Invariability and Adjustment on Pixel Pairs

Reversible Watermarking Based on Compensation

“…Both methods in [4] and [9] embed data by using pixel difference modification, which is similar with the proposed method. The superiority of the proposed method compared with [4] and [9] is thus mainly due to the compensation effect. The embedding capacity in Table 5 is the largest number of bits that can be embedded under the biggest modification of 1 to the pixel values of host image.…”

“…As can be seen in Table 5, the proposed method 2 has larger embedding capacity than the proposed method 1 for all test images. Our method can embed much more bits than [4] and with higher marked image quality. For example, the proposed method 1 can embed 48,731 bits into Lena with PSNR 50.83 dB, their method can only embed 38,562 bits with PSNR 48.82 dB.…”

“…Together with p u , around 50 percent of pixels stays unmodified after embedding data which can produce a peak signal-to-noise ratio (PSNR) value for marked image around 10 × log 10 (255 × 255/0.5) = 51.14dB. We then compare our method with [4] and [9] as shown in Table 5. Both methods in [4] and [9] embed data by using pixel difference modification, which is similar with the proposed method.…”

“…We then compare our method with [4] and [9] as shown in Table 5. Both methods in [4] and [9] embed data by using pixel difference modification, which is similar with the proposed method. The superiority of the proposed method compared with [4] and [9] is thus mainly due to the compensation effect.…”

“…Many improvements have been proposed based on DE and HS, such as prediction error expansion (PEE) [3], invariability and adjustment on pixel pairs [4], better location map construction [5], double-layered embedding and sorting [6], adaptive embedding [7], histogram modification of pixel differences [8,9], interpolation error histogram shifting [10] and the recently proposed two dimensional histogram shifting methods [11,12]. The above mentioned RW schemes are categorized as RW scheme without compensation because of the modification to host image can only be recovered after extracting data from the marked image.…”