Image Watermarking

Abstract

A method (100) of watermarking an image (101) is disclosed. The method (100) segments the image (101) into regions. Each region is then characterised based upon one or more statistical features of the respective regions. For each region, a watermarking method and a number of bits to be embedded into the respective region are selected based on the characterization of the region. The number of bits of data is then embedded into the respective region using the watermarking method selected for that region. A method (105) of decoding a watermarked image (145) is also disclosed.

Description

FIELD OF THE INVENTION

The present invention relates generally to image watermarking and, in particular, to adaptively watermarking an image by embedding a payload into the image.

BACKGROUND

The storing and distribution of media, such as images, is increasingly performed in a digital format. Unfortunately, the digital format also makes the copying of such media very simple whereby exact copies may be made, or alternatively for media to be altered without leaving obvious signs that the media has been altered. A need exists, therefore, for mechanisms to protect ownership of digital media, or to detect whether the media has been altered.

Digital watermarking is a technique that provides a solution to this need. Digital watermarks are pieces of information embedded to digital data, such as a digital image, that can be detected or extracted later to make an assertion about the data. The information can take many forms. The embedded information may be textual data about the author, its copyright, etc; or it may even be another image.

Digital watermarks are generally embedded into digital images in a manner that make the watermark invisible to a human observer as such watermarks do not cause degradation in the aesthetic quality, or in the usefulness of the images. The digital watermarks can be detected and extracted later to facilitate a claim of ownership, to extract relevant information, or to determine whether the image has been altered.

Watermarks may also be classified as ‘robust’ or ‘fragile’. Robust watermarks are those which are difficult to remove from the image in which they are embedded, and are therefore useful to assert ownership. Fragile watermarks are those that are easily destroyed by any attempt to tamper with the image. Fragile watermarks may therefore be used to determine whether an image has been altered, because if a watermark in a previously watermarked image is absent then it may be concluded that the image has been altered.

In recent years, medical images are also archived and communicated in digital form. For example, healthcare facilities and physicians use a picture archiving and communications system (PACS) to manage and archive patient images. Such images may also be distributed via the Internet. Another example of the increased mobility of medical images is the use of tele-radiology. Due to the sensitive nature of medical images, it is important that the authenticity of the images can be verified.

For reasons explained above, watermarking offers a solution to verify the authenticity of the medical images. However, it is particularly important with medical images that the perceptual integrity and quality of the images be maintained during the watermarking because the clinical reading of the images (e.g. for diagnosis) must not be affected. As a result the information carrying capacity of medical images is relatively low.

A need exists for a watermarking method which limits the visual degradation of images.

SUMMARY

A watermarking method is proposed for encoding a payload into an image in a manner which limits the visual degradation of the image.

According to a first aspect of the present invention, there is provided a method of watermarking an image, said method comprising the steps of:

segmenting said image into regions, with each region being a cluster of pixels within the image;

characterizing each region based upon one or more statistical features of respective regions;

for each of said regions, selecting one of a predetermined number of watermarking methods based on the characterization of the respective region;

for each of said regions, selecting a number of bits to be embedded into the respective region based on the characterization of the region; and

for each of said regions, embedding said number of bits of data into the respective region using the watermarking method selected for the region.

According to a second aspect of the present invention, there is provided a method of decoding a watermarked image, said method comprising the steps of:

segmenting said watermarked image into regions;

characterizing each region based upon one or more statistical features of respective regions;

for each of said regions, selecting one of a predetermined number of watermarking methods based on the characterization of the respective region;

for each of said regions, selecting a number of bits to be extracted from the respective region based on the characterization of the region; and

for each of said regions, extracting said number of bits of data from the respective region using the watermarking method selected for the region.

According to another aspect of the present invention, there is provided an apparatus for implementing any one of the aforementioned methods.

According to yet another aspect of the present invention there is provided a computer program product including a computer readable medium having recorded thereon a computer program for implementing any one of the methods described above.

Other aspects of the invention are also disclosed.

BRIEF DESCRIPTION OF THE DRAWINGS

One or more embodiments of the present invention will now be described with reference to the drawings in which:

FIG. 1 shows a schematic flow diagram of a method of watermarking an image to form a watermarked image;

FIG. 2 shows a schematic flow diagram of a method of decoding the watermarked image;

FIG. 3 is a schematic block diagram of a general purpose computer upon which arrangements described can be practiced; and

FIGS. 4A and 4B show an 8×8 block in the discrete cosine transform (DCT) domain used to explain the preferred DCT watermarking method.

DETAILED DESCRIPTION

Where reference is made in any one or more of the accompanying drawings to steps and/or features, which have the same reference numerals, those steps and/or features have for the purposes of this description the same function(s) or operation(s), unless the contrary intention appears.

FIG. 1 shows a schematic flow diagram of a method 100 of watermarking an image 101 to form a watermarked image 145 by embedding a payload 140 within the image 101. FIG. 2 shows a schematic flow diagram of a method 150 of decoding the watermarked image 145 to extract the embedded payload 140 from the watermarked image 145.

It is desirable that the perceptual error introduced through the watermarking of the image 101 be minimised. Limits on human visual detection of image degradation provide thresholds for tolerable image quality losses, which in turn provide an envelope for the amount of information which can safely be devoted to watermarking. One measure of perceptual error is a structural comparison providing a perceptual error metric. The structural comparison determines the amount of similarity between an original (unwatermarked) region and that region after watermarking, and is calculated as follows:

$\begin{array}{cc}SC=\frac{{\sigma }_{\mathrm{xy}}}{{\sigma }_x{\sigma }_y}& \left(1\right)\end{array}$

wherein σ_x, is the standard deviations of the original (unwatermarked) region, σ_y is the standard deviation of the watermarked region, and σ_xy is the correlation coefficient between the original and watermarked regions. Other measures may be used to estimate perceptual error.

The methods 100 and 150 of watermarking the image 101 and decoding the watermarked image 145 are preferably practiced using a general-purpose computer system 200, such as that shown in FIG. 3 wherein the processes of FIGS. 1 and 2 may be implemented as software, such as an application program executing within the computer system 200. In particular, the steps of methods 100 and 150 are effected by instructions in the software that are carried out by the computer system 200. The software may be stored in a computer readable medium. The software is loaded into the computer system 200 from the computer readable medium, and then executed by the computer system 200. A computer readable medium having such software or computer program recorded on it is a computer program product. The use of the computer program product in the computer preferably effects an advantageous apparatus for watermarking an image and decoding a watermarked image.

The computer system 200 is formed by a computer module 201, input devices such as a keyboard 202 and mouse 203, output devices including a printer 215, and a display device 214. The computer module 201 typically includes at least one processor unit 205, a memory unit 206, and an number of input/output (I/O) interfaces including a video interface 207 that couples to the video display 214, an I/O interface 213 for the keyboard 202 and mouse 203, and an interface 208 for the printer 215. A storage device 209 is provided and typically includes one or more of a hard disk drive, a floppy disk drive and a CD-ROM drive. The components 205 to 213 of the computer module 201 typically communicate via an interconnected bus 204 and in a manner which results in a conventional mode of operation of the computer system 200 known to those in the relevant art.

The methods 100 and 150 may alternatively be implemented in dedicated hardware such as one or more integrated circuits performing the functions or sub functions of watermarking an image and decoding a watermarked image. Such dedicated hardware may include graphic processors, digital signal processors, or one or more microprocessors and associated memories.

Referring to FIG. 1, the method 100 starts in step 105 where the processor 205 retrieves the image 101 from the storage device 209 and segments the image 101 into regions. In the preferred implementation the regions are 8×8 pixel blocks. However, other block sizes and block shapes may be used.

Also, the size of the regions may be made variable. In one implementation adjoining areas within the image that have similar statistical features (described below) are grouped to form regions of varying sizes. In a other implementation the areas within the image are classified according to content, such as texture or edges, and adjoining areas are grouped to form regions of varying sizes.

In step 110 the processor 205 then characterizes each region using one or more statistical features of respective regions. Such statistical features may include the standard deviation σ_R of the values of the pixel within the region.

Based on the characterization of each region, a watermarking method is selected by the processor 205 for each respective region in step 115. In a preferred implementation a selection is made between a spatial domain watermarking method and a frequency domain watermarking method. More particularly, the selection is made between the Least Significant Bit (LSB) and the Discrete Cosine Transform (DCT) watermarking methods known in the art. The LSB based watermarking method sets selected least significant bit(s) of pixels in a region to the binary values 0 or 1 based on a section of the payload 140. DCT based watermarking method sets selected least significant bit(s) of DCT coefficients of a region to the binary values 0 or 1 based on a section of the payload 140. By setting the least significant bit(s) of either pixels or coefficients the least amount of perceptual error is added to the region, and ultimately the image.

In the case where the standard deviation σ_R is used to characterize each region, then if the standard deviation σ_R below a predetermined threshold ø the LSB watermarking method is used. This is because the LSB watermarking method has been found through experimentation on 12-bit Magnetic Resonance (MR) images to produce less structural errors in the watermarked image 145 than the DCT watermarking method for lower standard deviation σ_R, with the structural errors being based upon the structural comparison SC presented in Equation (1). Thus, the LSB watermarking method introduces higher perceptual similarity for lower standard deviations σ_R than the DCT watermarking method. In the preferred implementation the predetermined threshold ø is set to 2⁵ for a certain type of MR images.

In the case where the standard deviation σ_R is above the predetermined threshold ø, then the DCT watermarking method is used. It has been found in the before mentioned experimentation that for higher standard deviations σ_R the LSB and DCT watermarking methods produce similar structural errors in the watermarked image 145. The DCT watermarking method is selected for higher standard deviations σ_R over the LSB watermarking method because the DCT watermarking method is more robust than the LSB watermarking method.

Next, in step 120, the processor 205 selects for each region a capacity level based on the characteristics of respective regions. The capacity level of any one region establishes the number of bits that may be embedded into that region without compromising the region's perceptual quality.

For certain examples of MR images the processor 205 preferably selects from 3 different capacity levels, which are 0, 1, and 2. That is, the processor 205 selects whether 0, 1 or 2 bits are to be embedded into the region. In the case where the regions are characterized using the standard deviation σ_R of each region, capacity level 0 is selected when the standard deviation σ_R is less than a first bound ε₁ because any changes made to pixel values at low standard deviations σ_R cause large structural errors. For standard deviations σ_R in a mid range, that is for standard deviations σ_R between the first bound ε₁ and a second bound ε₂, the capacity level 1 is selected because low levels of embedding is acceptable at such standard deviations σ_R. Finally, in the case where the standard deviations σ_R exceeds the second bound ε₂ structural errors are negligible, and the capacity level 2 is selected for such regions. For specific examples of MR images the first bound ε₁ may be set to 2³ and the second bound ε₂ may be set to 2⁵. The number of different capacity levels and associated bounds are predetermined for particular image types.

The method 100 then proceeds to step 125 where each region is embedded with a section of the payload using the watermarking method selected for that region in step 115 and the capacity level selected in step 120, thereby forming watermarked image 145.

Consider the case where an 8×8 block size is used for regions. Only 0, 1 or 2 bits are embedded in each region. Hence, in the case of the LSB watermarking method, only the least significant bit of 1 or 2 of the 64 pixels is changed. The preferred implementation uses the least significant bit of the first 1 or 2 pixels in the block to embed a section of the payload 140.

In the case of the DCT watermarking method, the high frequency coefficients are used to embed a section of the payload 140 as those coefficients introduce the least perceptual distortion to the image 101. FIGS. 4A and 4B show an 8×8 block in the DCT domain. The shaded block group 410 in FIG. 4A are the blocks used to embed 1 or 2 bits of the payload 140 in the block. In the case where 1 bit is embedded, all the block in group 410 are used, whereas when 2 bits are embedded the blocks in groups 420 and 425 shown in FIG. 4B are used to embed the 2 bits respectively.

A bit is embedded in a group 410, 420 or 425 by applying the exclusive OR (XOR) function to all the LSBs of the rounded DCT coefficients of the blocks (which are typically quantised) in the group 410, 420 or 425. If the result of applying the XOR function is the same as the value of the bit to be embedded, the DCT coefficients of the group 410, 420 or 425 remains unchanged. If the result of applying the XOR function is different to the value of the bit to be embedded, the value ‘1’ is added to the rounded DCT coefficients of the blocks of the group 410, 420 or 425, such that the smallest possible perceptual error occurs. This process is repeated until the result of applying the XOR function to the LSBs of the rounded DCT coefficients of the blocks in the group 410, 420 or 425 is the same as the the value of the bit to be embedded.

Let b_p be the number of bits in the payload 140 and b_m be the sum of the capacity levels of all the regions. If b_p<b_m, that is the payload 140 is less than the total payload capacity calculated for the image, then the payload 140 is embedded in the first number of regions until all the sections of the payload 140 are embedded. Preferably the payload 140 is relatively small. As a result b_p<b_m lo would typically apply for most images.

If b_p=b_m, that is the payload 140 exactly matches the total payload capacity calculated for the image, then the payload 140 is embedded in each region. If b_p>b_m then the payload 140 exceeds the total payload capacity calculated for the image. In this case the first b_m bits of the payload 140 is embedded.

Reference is now made to FIG. 2 wherein the schematic flow diagram of the method 150 of decoding the watermarked image 145 to extract the embedded payload 140 is shown. The method 150 starts in step 160 where the processor 205 retrieves the watermarked image 145 from the storage device 209 and segments the watermarked image 145 into regions in a manner corresponding to the manner in which the image 101 was segmented in step 105 (FIG. 1).

The processor 205 then in step 165 characterizes each region in a manner corresponding to the maimer in which the regions were characterized in step 110 (FIG. 1). Based on the characterization of each region, a watermarking method is selected by the processor 205 for each respective region in step 170 using the same criteria for selection as was used in step 115. The character of the regions would remain unchanged following the embedding of a section of the payload 140 in each region because the very little change is introduced in each region. Accordingly, as the same criteria are used in step 115 (FIG. 1) and step 170, the watermarking method selected would be the same.

Step 175 follows where the processor 205 selects for each region a capacity level based on the characteristics of respective regions, again in a manner corresponding to the manner in which the capacity level was selected in step 120 (FIG. 1).

Finally, the method 150 ends in step 180 where a section of the payload is extracted from each region based upon the watermarking method selected for that region in step 170 and the capacity level selected in step 175. Sequentially appending the respective sections of the payload, the payload 140 is formed.

In one implementation an end-of-payload special marker, which forms part of the payload 140 embedded in the image 101 during method 100, signifies the end of the payload 140. In another implementation the payload 140 has a predefined bit-length b_p. In yet a further implementation the method 150 processes all the regions as set out above. As sections of the (real) payload 140 are typically only embedded in a first number of regions, the sections of the payload extracted from the remainder of the regions would include noise. A user may then make a determination where the real payload 140 ends.

It is noted that the method 150 of decoding does not require side information, such as the number of bits b_p in the payload 140. This is particularly useful for medical images due to the fact that it is desirable to only embed a relatively small payload into medical images. Large payloads have a particularly detrimental affect of medical images for reasons explained in the “Background” section.

Methods 100 and 150 are also particularly useful as the original (unwatermarked) image 101 is not required to be available when decoding the watermark image. The payload can be extracted from the watermarked image 145 only.

The foregoing describes only some embodiments of the present invention, and modifications and/or changes can be made thereto without departing from the scope and spirit of the invention, the embodiments being illustrative and not restrictive.

Claims

1. A method of watermarking an image, said method comprising the steps of: segmenting said image into regions;characterizing each region based upon one or more statistical features of respective regions;for each of said regions, selecting one of a predetermined number of watermarking methods based on the characterization of the respective region;for each of said regions, selecting a number of bits to be embedded into the respective region based on the characterization of the region; andfor each of said regions, embedding said number of bits of data into the respective region using the watermarking method selected for the region.

2. A method according to claim 1, wherein at least one of size and shape of said regions is variable dependent upon one or more properties of pixels in respective regions.

3. A method according to claim 1 or 2, wherein said one or more statistical features include the standard deviation σR of the values of pixel within respective regions.

4. A method according to claim 3, wherein said number of bits is selected dependent upon predefined ranges of said standard deviation σR.

5. A method according to claim 1, wherein a selection is made between a spatial domain watermarking method and a frequency domain watermarking method.

6. A method according to claim 5, wherein said spatial domain watermarking method is the Least Significant Bit (LSB) watermarking method.

7. A method according to claim 5, wherein said frequency domain watermarking method is the Discrete Cosine Transform (DCT) watermarking method.

8. A method according to claim 5, wherein said one or more statistical features include the standard deviation σR of the values of pixel within respective regions and wherein said spatial domain watermarking method is selected when said standard deviation σR is below a predefined threshold, and said frequency domain watermarking method is selected otherwise.

9. A method of decoding a watermarked image, said method comprising the steps of: segmenting said watermarked image into regions;characterizing each region based upon one or more statistical features of respective regions;for each of said regions, selecting one of a predetermined number of watermarking methods based on the characterization of the respective region;for each of said regions, selecting a number of bits to be extracted from the respective region based on the characterization of the region; andfor each of said regions, extracting said number of bits of data from the respective region using the watermarking method selected for the region.

10. A method according to claim 9, wherein at least one of size and shape of said regions is variable dependent upon one or more properties of pixels in respective regions.

11. A method according to claim 9 or 10, wherein said one or more statistical features include the standard deviation σR of the values of pixel within respective regions.

12. A method according to claim 11, wherein said number of bits is selected dependent upon predefined ranges of said standard deviation σR.

13. A method according to claim 9, wherein a selection is made between a spatial domain watermarking method and a frequency domain watermarking method.

14. A method according to claim 13, wherein said spatial domain watermarking method is the Least Significant Bit (LSB) watermarking method.

15. A method according to claim 13, wherein said frequency domain watermarking method is the Discrete Cosine Transform (DCT) watermarking method.

16. A method according to claim 13, wherein said one or more statistical features include the standard deviation σR of the values of pixel within respective regions and wherein said spatial domain watermarking method is selected when said standard deviation σR is less than a predefined threshold, and said frequency domain watermarking method is selected otherwise.

17. Apparatus for watermarking an image, said apparatus comprising: means for segmenting said image into regions;means for characterizing each region based upon one or more statistical features of respective regions;means for selecting, for each of said regions, one of a predetermined number of watermarking methods based on the characterization of the respective region;means for selecting, for each of said regions, a number of bits to be embedded into the respective region based on the characterization of the region; andmeans for embedding, for each of said regions, said number of bits of data into the respective region using the watermarking method selected for the region.

18. Apparatus for decoding a watermarked image, said apparatus comprising: means for segmenting said watermarked image into regions;means for characterizing each region based upon one or more statistical features of respective regions;means for selecting, for each of said regions, one of a predetermined number of watermarking methods based on the characterization of the respective region;means for selecting, for each of said regions, a number of bits to be extracted from the respective region based on the characterization of the region; andmeans for extracting, for each of said regions, said number of bits of data from the respective region using the watermarking method selected for the region.

19. A computer program product including a computer readable medium having recorded thereon a computer program for implementing the method of claim 1.

20. (canceled)

21. (canceled)

22. A computer program product including a computer readable medium having recorded thereon a computer program for implementing the method of claim 9.