TW201023587A - Method of high capacity reversible data hiding scheme using edge prediction error - Google Patents

Abstract

A high-capacity reversible data hiding method is presented. It is based on prediction-error expansion. To increase the embedding capacity, an edge-based predictor is proposed to achieve better prediction performance for the pixels which located at the edge of the image. Experimental results show that the PSNR of the embedded image by the proposed method is 1dB higher than Thodi's method under the same hiding capacity. By applying the embedding method to an image more than once (multiple embedding), the proposed method achieves higher embedding capacity (about 2.6bpp) than Tian's scheme. There is also a significant improvement in the quality of the embedded image.

Description

201023587 Description of the Invention: [Technical Field] The present invention is a method for hiding information into images, in particular, a recoverable information hiding method having a high hidden amount. About [Prior Art] Data hiding is a way to hide confidential information in video technology. By transmitting information with hidden information to achieve confidential information and image transmission.

Φ J M prevents the image from being improperly copied. P is a method of correcting the content, tracking the order, distributing the information, and monitoring the information. After the information hiding method of the common Shennu is hidden, the human eye can't distinguish the difference between the hidden image and the original image, but it is a permanent distortion to the image. In some applications, especially medicine. Imaging images for military use 'requires the preservation of completely correct images, even small differences are not allowed. Therefore, some scholars have proposed that some methods can be restored as images hidden in the medium. This method is called reversible data hiding. In addition to the general two-dimensional image, you can use the recoverable data hiding method. In the audio, 3D video and video, for example, the relevant information is hidden in the audio, 3D video and video, as supplementary information, increase the signal content, and identify The correctness of the content or whether the signal is destroyed or not. In recent years, many scholars have proposed different recoverable data hiding methods, which are related to the research on the amount of images and the quality of the images after storage, such as the statistical map displacement method proposed by scholars such as Ni. (histogram shift) [Zhicheng Ni ' Yun-Qjng Shi > Nirwan 201023587

Ansari, and Wei Su, “Reversible data hiding,” IEEE Trans, on Circuits and Systems for Video Technology, vol. 16, no. 3, pp. 354 - 361, Mar. 2006], whose information is about 〇 .3bpp(bits per pixel) and the PSNR (Pixel Signal to Noise Ratio) quality of the image after hiding is 48dB; the difference expansion method proposed by Tian [丄Tian, “Reversible Data Embedding Using a Difference Expansion” , IEEE Trans, on Circuits and Systems for ❹ Video Technology, vol. 1 3 > No. 8 > pp. 890-896 > Aug. 2003. ; J. Tian and Raymond O. Wells, Jr., “ Reversible data-embedding with a hierarchical structure ” , International Conference on Image

Processing (ICIP), vol. 5, pp. 24-27, Oct. 2004] has information that can be hidden more than 2bpp, and the PSNR quality is about 15dB.

After the method proposed by Tian has hidden information of about i.5bpp or more, the image is like a spurious noise, and the distortion is serious. Although the method proposed by Tian has the advantage of being able to hide a large amount of data, its quality is not good. Therefore, there is no effective control law for users to obtain the best image quality depending on the amount of storage. The problem in this area has been the subject of many research's such as the histogram shift of Thodi et al. [D. M. Thodi and J. J. Rodriquez, “Predict-error-based reversible watermarking,’, in

"Currently, Rep. - Expansion embedding techniques for reversible • watermarking > IEEE Trans, on Image Processing > vol. 13' no_ 3, pp. 721- 730, Mar· 2007.] is an example.

In addition to the proposed histogram shift to improve the inventory control mechanism, Thodi also proposes a prediCti〇n-error 〇 expansion method that reduces the image distortion after the information is hidden. The prediction error expansion method is the difference between the predicted value and the target value. As the medium with information, if the estimated value is correct, the amount of information that can be hidden is higher, and the quality of the hidden image is better. Therefore, the prediction error The expansion method 'has a large amount of information hidden in the whole' and the image distortion after hiding the information is small. However, the method of Thodi requires that the trough head (overheac|) is larger than that proposed by Dingjail. If it is tested under the control of the statistical displacement method, it is recommended by Thodi when the amount is below 0.2bpp. The image quality is slightly worse than Dingjan's method. ® Thodl et al.'s prediction error value expansion method uses the correlation in the image to obtain the difference between the predicted value and the original value as a medium for hiding data, and the expansion operation doubles the difference. Then, the hidden assets are placed in the last bit that becomes null, so when the prediction result is closer to the correct value, there is a smaller error, and the better the quality after hiding, the available hiding area is also more. Observing the location map of Thod丨 et al., it is found that the probability of being hidden in the boundary area is high. The so-called location map is used to indicate whether the pair of pixels in the original image (Pairs of pixe|s) is expandable, that is, 201023587

Thodi is a location map of whether or not it can be hidden. The prediction error value expansion method is further explained below: (1) Prediction rule: The method proposed by Bishan et al. scans all the pixels in the image in the order of laser sweep (four) for prediction. Please refer to the fifth figure, ❹ three predicted reference values, heart, heart, h to predict the value a. The order of prediction is from the left so that the predicted value does not refer to the reference value that has information hidden. == The prediction order used in the restoration is the reverse order of the Tibetan. The U-method j and the predicted value used by Thcan et al. are shown in equation (1). 3 of them are predicted values. Max(c2, c3), if c, < min(c2, c3) min(C2,Cj),ifCi>max(C7,q) ,c2 +c3 otherwise

Another method of predicting hiding proposed by Th〇d丨 et al. is shown in the following formula (2). The difference between the measured value and the target value is h, which is the medium for Tibetan information. Hiding money. The ft method is an extended operation, and h is the difference result h after the expansion operation. The result is defined as the difference in which the information has been hidden. a, is the pixel value of the information that has been hidden. h = ",h' = 2h+b,(2) ^ where b is the information value. For the 8-bit image, the value of the pixel used to hide the beacon must be between 255 and 255. , can be extended, otherwise, do not do any operation on a. Thidj et al.'s predictive reduction method, as shown in the following formula (3). The steps include: first calculate the difference h' of hidden information, restore 201023587

Difference ί, still live, pixel value and extract Tibetan information h’ = a’ - S

^ b = h' mod 2

<—* /N a = a + h (3) 能否 Whether the pixel pair listed above can be expanded and recorded with a location map of the same size as the original image. This map is necessary for restoring the image. Therefore, you need to hide this map into the image. In order to reduce the size of the hidden data, we compress the map in JBIG2 format. (2) Statistical 囷 Displacement Rule: In order to improve the control mechanism of the volume control, Thodi et al. proposed a statistical displacement method to improve the image quality of the Tiari method when the amount is small. Ding|When the method of grasping Q is small, it is still possible to expand the larger difference, so the image quality is not good. Thodi's statistical displacement method takes a critical value for the expandable difference. If the absolute value is smaller than the critical value, the expansion is performed. The area larger than the critical value is displaced. Let the value fall in a larger or smaller area. It is recognized when decoding. The chart displacement method is described in detail as follows: Please refer to the sixth figure, which is the statistical result of the extended difference of the 512x512 8-bit gray-scale image. The horizontal axis is the difference value and the vertical axis is the quantity. The expansion operation is performed by selecting a part of the value in the chart according to the size of the space to be hidden, wherein the value of the small difference has the highest occurrence rate. 8 201023587 Please refer to the seventh figure' In order to reduce the gap between the images after hiding, the value of the relatively small absolute value is preferred to obtain an optimal threshold. Selecting a critical value > 〇 ‘ for the combination of the difference within the range is the 丨nner region in the seventh graph. After the expansion, the range of the value falls to [-242d + l] ', which is |nner regj0ri, so only the value originally in the range of [-4J] is a range that can be hidden. Shifting the other parts of the Outer region to the left or right, that is, subtracting J from the difference of negative numbers and adding +1 to the difference of positive numbers to avoid the extended value Opportunities for overlap, as shown in Figure 8. The calculation and reduction operations of the displacement are as shown in equations (4) and (5). Hh + A+ \, ifh>A h~ Δ, if h<-Δ (4) h [VA-l,ifhs>2A + l hs+A, ifh <-2Δ (5) ~ where λ is not hidden The difference of the incoming information, \ is the difference after the displacement, and is the difference after the displacement calculation. As described above, the advantages and characteristics of the statistical displacement method include: (丨). Since the address corresponding to the graphic is fixed under any quantity, it can be pre-calculated by adding 'plus The secret information (p_ad), you can find the optimal threshold value from the statistical map to achieve the storage control; (丨丨). The displacement part does not need complex calculations and the displacement is small when the storage is small. The amount is not large, but it can effectively make the 9 201023587 expansion calculation with a small difference, so it can be used to make a good image quality. Cattle*. For the hidden operation, please refer to the ninth figure, the area value of the fake implementation, the lower right and the lower ^ ^ ^ pixels are the predicted value a, the other areas are the predicted values, and the value Mi is hidden. Therefore, the hidden operation is as follows: expected test (1)·calculate the predicted value, calculate the predicted error value 58 a = c2+c3-Cl=56 + 60-58 〇^ = ^-^ = 55-58 = -3 benefit (8) · Determine whether Λ is less than the critical value. If it is smaller, do the expansion operation and hide the information. Otherwise, do the displacement operation. (丨丨丨·)· Do the expansion operation and get the difference between the hidden information, ,. Calculate the pixel value of the hidden information /z' = 2x/j + 6 = 2x(-3) + U—5 Ο α' = λ + // = 58 + (~ 5) = 53 If the restored pixel value is Between [〇, 255], it is judged that it can be legally hidden, and if it is not, the expansion operation is not performed. After predicting each pixel value of the image and hiding it, you can get an image with a video. In addition, for example, the Th (10) fetch information and the restored image, please refer to the tenth®, the lower right area is the predicted value of other areas with the information value, and the information extraction and image restoration are calculated as follows: 201023587 (i) Calculate the predicted value S ' to calculate the predicted error value V. ^ = 56 + 60-58 = 58 53~58 = ~5 (ii) The difference is less than 2 times the critical value, which means that there has been an expansion operation and a value is hidden.

(iii) The reduced difference stone 'takes out the hidden value f and restores the pixel value α. Hf -5 X _ 2 _ =—3

a = a + h = 58 — 3 = 55 6 = /7, mod2 = l (iv)· It is known from the restored values β = 3 and = f that this method can correctly retrieve the information and completely restore it. Based on the above description, the flow chart of the information hiding method proposed by Thodi can be shown in the eleventh figure. The hiding steps are briefly described as follows: (i)·calculate the prediction error; (Η)_Extended judgment, the test can be (iii), according to the test result of the scalable quantity, record the position map, 201023587 and compress the graph; (iv). Combine the compressed position corresponding graph with the secret information (pay|〇ac|) * Hide the string 'select the appropriate threshold value from the length of the hidden string; (v) . Execute the critical value judgment, when the judgment result shows that the expansion operation is performed, the data string to be hidden is hidden, and at the same time Completion of the displacement operation; (vi) The position map and the value of the threshold value overwrite the pixels in the image by a LSB replacement. In order to achieve the reversible nature, the overwritten bit must be hidden in the extended location after the payload. The hidden string indicates the twelfth A and B pictures. Figure 12A is a schematic diagram of the hidden information content, which indicates the state after the secret information (pay丨〇ad). The latter half needs to reserve some space first (empty); the length of the hidden bit content It is the same as the position-corresponding graphic (as shown in Figure 12B), and since the position-corresponding graphic is replaced with a minimum bit (LSB rep is hidden in the first few pixels of the image, it is possible to completely restore the image, ❹ Write the replaced bit (overwmebit) into the reserved space. In front of the position map, 'add a few information, including the length of the position corresponding graph, the size of the displacement threshold, and the length of the secret information. When restoring, the above information is decoded first to complete the decoding. After the information is hidden, the obtained image is the shadow of the secret information. When decoding, the information in the image can be directly used to obtain the hidden information. Incoming information and restoring images, no need to transmit additional information. In addition, the method of information restoration is hidden, and the steps are briefly described as 12 201023587 (丨). From the image Corresponding to the pixel pattern in front of the removal position and decompression (Shushu) is covered by reducing bit (overwrite bit) -. (Πί) · extracted secret information (payload) 〇ν) _ reduction extension (Expansion) and displacement (Shift) operation. In other words, the method of information restoration is the reverse execution step of the information hiding method. According to the foregoing description, the method provided by Thodi et al. has a high probability that the boundary region cannot be hidden in information, resulting in an overall information hiding amount of low. . SUMMARY OF THE INVENTION In order to solve the problem that the existing information hiding method has a high probability that the boundary region cannot be hidden, resulting in a low amount of information that can be stored, the present invention adopts a predictor that can consider the change of the graphical boundary to The information in the boundary area is more correctly estimated, and the image quality and the larger amount of storage are better obtained.配合 In combination with the foregoing technical problems and the object of the invention, the present invention provides a high-capacity recoverable information hiding method using edge prediction error, the steps comprising: predicting an error value, selecting an original image for each pixel of the original image Performing an error value prediction 'which specifies each pixel as a prediction target' generates a prediction error value by a three-stage edge-guided predictor (edge_based predictr), the three-stage predictor comprising: One-half the number of pixels in the image respectively perform error value prediction 'the pixels whose number of columns of the original image is even, the number of rows is odd, or the number of rows is odd, and the number of columns is even is designated as the 13 201023587 predicts the target' and uses the four neighboring pixels of the upper and lower left and right of the predicted target as the predicted reference value, and calculates the difference between the predicted reference value and the predicted target, which is determined according to the following predicted value setting judgment method. The predicted value, the step of determining the calculation method of the predicted value includes: (1) if the four predicted reference values are equal, directly The predicted reference value is set as the predicted value of the predicted target; (2) if any two predicted reference values are equal, and the predicted reference value with the same value is set to the predicted value; and 〇(3) if four predicted reference values If they are not equal, then the two values with the numerical value in the middle are selected and averaged, and the average value is the predicted value; ❹ the first Is self segment. Pre-, the first quarter of the original image The J system uses the pixels whose original image number is odd and the number of rows is odd as the prediction target, and then takes the upper left, lower left, upper right, and right of the predicted target: the values of the four neighboring pixels are the predicted reference values. And using the predicted value " again to determine the 4 calculation method to calculate the predicted value of each pixel; and, the third stage: the last remaining quarter of the original image of the pixel will be predicted Pixels whose number of columns is even and whose number of rows is even are measured one by one, and the predictions of each pixel are calculated for each pixel of the upper, lower, left and right sides of each prediction target. And, among them, At the end of the stage - a bit directly performs the expansion = boundary value judgment, which is based on Thodi's prediction error expansion method image, lj break and - critical value judgment 'the extension judges the mother-pixel of the original scene Whether the difference between the predicted value and the target value can be performed 14 201023587 - Extended operation, if the result of the break is i, the critical value judgment is performed, and vice versa, and the critical value judgment determines the absolute value of each pixel The relationship between the value and the -threshold value, when the absolute value is less than the critical value, an expansion and hiding are performed, and when the absolute value is larger than the critical value, the operation step is performed, wherein: 1 # The expansion and the hiding step are poor The pixel whose value can be expanded is used to hide the information action; and the step of shifting the operation is performed by accumulating the value of the non-expandable pixel and causing the pixel value to be displaced. The 'expanding and hiding step' uses one of Tjan's multi-layered hiding methods to hide information into the pixels of the original image. The expansion and hiding steps are first carried out by using one of the Tjari multi-layer hiding methods to hide the information of the first to the second layer, and then the information of the original image is hidden by the method of Thod. Wherein the critical value is selected by one of Thodi's statistical map displacement methods. Thereby, the present invention can hide information in the original image with the boundary, and greatly enhance the information hiding amount. [Embodiment] Please refer to the first figure, which is a preferred embodiment of the high-capacity recoverable information hiding method using edge prediction error according to the present invention, the steps of which include executing - a known edge-oriented predictor (1 〇), Expansion and threshold judgment (2 〇) as well as expansion, hiding and displacement (30). The implementation of the three-stage edge-directed predictor (1 〇> step: in order to improve the method proposed by Thodi et al. to achieve a better boundary pre-test 15 201023587, the preferred embodiment proposes a new prediction rule, the purpose of which is to image The correlation is best utilized. Refer to the method commonly used in the compression method, in the four directions of the target pixel's up, down, left, and right directions or the top left, bottom left, top right, and bottom right. As a prediction reference, if two of the four predicted values are the same, then the boundary is assumed to be the boundary and the predicted value is set as the boundary value. The three-stage edge-oriented predictor (1〇) step of the implementation is one or three. The edge-based predictor (edge_based predictr) predicts the error value for each pixel in an original image that is to be hidden into a secret information. The © predictor performs three stages, in accordance with a prediction in each stage. The four neighbor pixel characteristics of the target are divided into several methods for generating predicted values. The stages performed by the three-stage edge-directed predictor are described as follows: Stage--: Divided into the original image - the number of pixels to be predicted, the number of columns of the original image is even, the number of rows is odd or the number of rows is odd, the number of columns is even number is set as a prediction target, such as the second A map and the third = In the figure: A: The pixel values of '(M), (G2), (Q4), (11), (13), etc., which are not "1", are the predicted targets, 〇, 〇) The pixel values of the blocks of (12), © ( '4) (3'G), (3, 2), etc. are - prediction reference values. Referring to the second B picture, the upper, lower, left, and right four of each prediction target (4) are taken: the value of the neighboring pixel is the prediction reference value (6) to C4) to the prediction target.

Ma) - Predicted values are predicted. The calculation method of the predicted value is mainly based on the boundary area of the reserved image _, and the setting method of the value may include the following conditions: I(1) The right four prediction reference values (C1~C4) are equal, and the pre-1' is directly The test value (C1 to C4) is set as the predicted value of the predicted target (a); * If there are 4 predicted reference values (C 1 to C4) equal, it is assumed that the original 16 201023587 sets the boundary value as the pre-image There is a boundary passing the predicted target value; (3) if the four predicted reference values (C1~C4) are not equal, the four reference values (C1~C4) are sorted by size, and the value is selected in the middle. The two values are averaged, and the average value is defined as the predicted value. The second stage: predicting the quarter-number of pixels in the original image, the number of columns of the original image is odd, and the number of rows is odd. The pixel is used as the prediction target as shown in the third A diagram and the third b diagram. As the first-order strict method, the number of images marked with "2" in the third A picture is used as the prediction target, and the values of the four neighboring pixels of the upper left, lower left, upper right, and lower right of the measurement target are taken as prediction reference values. The positional relationship between the predicted value and the predicted reference value is as shown in the third B-picture, where 3 indicates that the predicted target Ί, ... 4 is the predicted reference value, and the predicted value of the second stage is calculated as the first-stage.

The third stage: predicting the last remaining quarter of the pixels in the image, using pixels with even numbers and even numbers as prediction targets, as shown in the fourth and fourth graphs. The pixel of the "3" indicated by the fourth A icon is used as the prediction target, and the four neighbors of the upper, lower, left, and right of each prediction target are used as the gastric reference value. The positional relationship between the predicted value and the predicted reference value is as shown in the fourth B-picture, where a indicates that the target is not predicted, and the heart, q, heart, and heart are predicted reference values. ^ The information generated in the third stage is the first part to be decoded and judged when the hidden information is retrieved from the image decoding in the future. In order to prevent the predicted value of the third stage from being affected by the hidden information, it will be the third. The last bit (LSB) of the predicted reference value of the phase is not included and is directly set to zero, so that the predicted value obtained by the hidden and the extracted end is the same, as shown in the following formula (6), where 〆, 17 201023587 second stage Calculation of predicted values (6) 4, C; and C: The actual predicted reference value is the same as the first and second stages. c; = c, -c, mod2 c'2 = c2-c2 mod 2 C3 = c3-c3 mod2 c; =c4-c4 mod 2 with f, mod2 table This extended operation and judgment (20) steps: current column The implementation of the three-stage edge-directed predictor (10) step completes all the prediction error expansion methods in the original image according to the three six segments, and the prediction is performed for three pixels, respectively. The completed prediction target is directly expanded and migrated (GVerhead) only to the third stage of the third-order (10) step. Since the preferred embodiment 1 uses the boundary error calculation, it is possible to obtain a smaller difference, a larger number of bins & and a better image quality. In addition, if the broadcast is larger than the third stage, then the :::: kind of situation often occurs in a small area, about ~ In order to improve this shortcoming, we will change the forecast stage to only the stage. The predictions and Tibetans, to meet the requirements for small reserves and high image quality. The prediction boundary value judgment (2G) step is based on the extension method proposed by ThGdi, and first determines the difference between the estimated value, the target value, and the estimated and target values for each pixel in the original image to determine the pixel. Difference 18 201023587 It is possible to expand 'When the judgment result is "Yes", the critical value is judged. $成 expandable (4) (4) to carry out the threshold value, which is the value of the extension - the threshold value, when the absolute value is less than the critical value, the -expansion and hiding (32) steps are performed, and the pixel area larger than the critical value is executed. A displacement is carried out in the (34) step, allowing the value to fall in a larger or smaller area to enable decoding: identification. The expansion, hiding, and shifting (30) steps, when the foregoing step (4) is completed, a multi-layer hiding method is proposed to improve the purpose of hiding data into the original image, and the method of multi-layer hiding is first proposed. The method is hidden. If the multi-layer hiding method cannot be used, the method of hiding the partial (1) is used instead. : After the image is hidden once, the correlation between adjacent pixels is reduced. After a large amount of information is reduced, the image quality will drop rapidly, and the size of the slot head (. state head) is limited by the proposed method. More, therefore, in this step (4), the first layer of Tian is hidden in the ancient, the soil, the society ^ ^ Μ Tibetan method to carry out the pre-~: layer of information hidden after the 'then, finally completed by the method of partiality The last-time information is hidden. ... The information carried out by the preferred implementation steps can be used to allow the original image to complete the two- to three-layer information Tibetans, with a storage amount of 166~2 62_, and an image quality of about 18 dB PSNR. In order to verify the effect of the preferred implementation method on the actual data hiding, please click "Annex 1 ~ 4" for four different content, the image size is 512x512 pixels and is 8-bit grayscale image;豕,八中' We use JPEG compression: 90% of Lena images as hidden information, and with this preferred implementation: method - layer and three layers of information hidden in 'and at the same time compare Ding (4) proposed Information Tibetans. Among them, (8) (8) distracted original images and images of Tibetans completed by 19 201023587 = Shi, and (4) (d) represent graphs comparing the preferred methods with the hidden features of the Th0di method. Attachment 2, the results of the comparison with the Annex I are performed in different schemas. 〃 We have repeatedly added the proposed information hiding method to achieve high capacity. The information hiding method is to use the image. Boundary changes are predicted. When the information is hidden, the 'boundary prediction method will gradually become unusable, so when you are hiding in the eve*, you will try to use the saga (1) and others after you are unable to use the proposed method. The method of Tibetans. The proposed high-capacity recoverable information hiding comparison object is the information hiding method proposed by Tian, and the method can reach the capacity close to the foot p. We use each direction to interact with different directions for Tibetans, such as "the first layer to level The direction takes two or two—the group is Haar converted, and the second layer uses the vertical direction to take two or two sets for Harr conversion. In order to verify that the proposed retrievable information hiding method has high capacity, we use four 8-bit grayscale images of size 512x512 m Λ h and about Z for airp丨ane, Lena, Peppers, Bab〇on, The experimental results are as shown in Annex ❹ V to Annex 12, where the visual quality of the images in Annex 5 to the man-made information is compared, and the annexes 9 to 12 are based on the multi-layered (4) _ embedd丨ng algorithm. The experimental results of the method proposed by Xiaoding & In the best case, the proposed method can hide three layers, and the information hidden in the file can be up to 2.62_. The image quality is like the noise, but the important image information in the image can still be recognized. The PSNR is the worst. In (5) already. From the comparison in the figure, it can be clearly seen that the proposed method is more than the original method of the differential expansion (difference expansi〇n). The lower image product f is preferably about fine or even 20 201023587 10dB 〇 We can see from the pictures and data hidden in multiple layers that the more data is hidden, the image quality is gradually declining. For different images, the more complex the image changes, the less the amount of images can be hidden. For example, the attached coffee has complicated hair. In the experiment, only two layers can be hidden, and 1.33bpp is obtained. The amount of image and PSNR his (3) day image quality, but the change of _ _ five can be hidden up to 2.62bpp, image 18.15dB. BRIEF DESCRIPTION OF THE DRAWINGS The first figure is a flow chart of a preferred embodiment of the present invention. A second A, B ® & schematic diagram of a first predicted target and a predicted reference pixel relationship of a preferred embodiment of the present invention. The third A and B diagrams are schematic diagrams of the second prediction target and the predicted reference pixel relationship in the preferred embodiment of the present invention. The fourth and fourth graphs are diagrams showing the relationship between the third prediction target and the φ prediction reference pixel in the preferred embodiment of the present invention. The fifth figure is a schematic diagram of the predicted target and predicted reference pixel relationship of the existing ThocN method. The sixth picture shows the pixel error statistics of the existing Thodi method. The seventh figure is an expanded pixel error map of the existing Thodi method. The eighth figure is the expanded and shifted pixel error statistical graph of the existing Thodi method. The ninth figure is a schematic diagram of the hidden operation calculation of the existing Thodi method. 21 201023587 The tenth figure is a schematic diagram of the restored image calculation of the existing Thod i method. The eleventh figure is a flow chart of the information hiding method of the existing Thod i. The twelfth A and B pictures are schematic diagrams of the information content of one of the existing hidden strings. '[Main component symbol description] None.

twenty two

Claims (1)

201023587 VII. The scope of application for patents: 1 · A method for using the edge prediction error, the steps of which include: A-recoverable information concealing the prediction error value, _--the original image performs an error value pre-μ raw per pixel The target of the image, the bean is _ two b,,, and each pixel is determined as a prediction, and the second positive segment edge-guided predictor generates: r-box "value, the stage of the operation performed by the predictor includes - Predictive error ❹ In the first stage, the prediction of the execution error value of the binary image in the original image is such that the number of pixels of the quantity is odd or the number of lines is odd; * the number of columns of the image is even, and the number of lines is a number For the odd-numbered, even-numbered pixel prediction target 'and the prediction target', 曰疋 is the pre-formed beam to the left and right four neighboring pixels as the prediction reference value, and the prediction is used as the basis for the following - prediction Value setting judgment leaf == target calculation difference 'Predictive value setting judgment The calculation method includes:; determining the predicted value, (1) if the four predicted reference values ❹ value is white, the line is straight The predicted reference value is the predicted value of the predicted target 丨^ = if any two predicted reference values are equal' and the predicted reference value having the same value is set to the predicted value; and (3) if four predicted reference values Λ If you are not interested, select the two values of the numerical value to average the average of the predicted values; predict the third segment: perform a quarter of the pixels in the original image, and set the number of original images as The number of odd-numbered pixels with an odd number of rows is 7 (the fourth target), and the values of the upper left, lower left, upper right, and lower four neighboring pixels of the prediction target are taken as the prediction reference value, and the prediction value of 23 201023587 is utilized. The sixth calculation method is used to calculate the predicted value of each pixel; and the third stage is to predict the last remaining quarter of the image in the original image, and the number of columns is even and the number of rows is even. The pixel is used as the prediction target, and the four neighbors of the upper, lower, left and right of each prediction target are predicted reference values', and the prediction calculation method is used to calculate the calculation method for each pixel. The measured value, wherein the last bit of each prediction reference value at this stage is directly set to zero; and the expansion and the threshold value judgment are performed by one of the prediction error expansion methods of Ding H〇di, and the expansion judgment and the _threshold value are performed. Judging, the extension determines whether the difference between the measured value and the target value of each of the original images can be expanded. If the right judgment result is yes, the threshold value is judged, and vice versa. The threshold value judges the relationship between the absolute value of the difference value of each pixel and the -threshold value, and performs - expansion and hiding when the absolute value is smaller than the critical value, and performs a displacement when the absolute value is larger than the critical value. The operation step, wherein: the expanding and hiding step is to perform a full Q input information action on the pixel whose difference value can be expanded; and the step of shifting the operation is to perform a value incrementing of the non-expandable pixel to cause the pixel value to be displaced. . 2. The high-capacity recoverable information hiding method using edge prediction error as described in the scope of patent application. The expansion and hiding step uses the Tian-multilayer hiding method to hide information into the image of the original image. in. The step of the edge prediction error is first used. 3. The capacity-recoverable information hiding method described in item 1 of the patent application scope, the expansion and the hiding method of a multi-layered hiding method of 24 13⁄4 201023587 Tian-- The information of the layer is hidden, and the information of the original image is hidden by using the method of Dinghodi. 4. A high-capacity recoverable information hiding method using edge prediction errors as described in claim 2 or 3 of the patent application, the threshold value being selected by a Thodi chart displacement method. Eight, schema: (such as the next page) ❿ 25