Disclosed is a system and method for determining preferred
image locations for placing watermark information for both hidden and obvious marks,
and more particularly the use of at least one of three criteria (image similarity,
contrast, and image value range) to determine the best area of hiding or showing
the watermark, depending upon the user's intent for the mark.
BACKGROUND AND SUMMARY
It is known to use digital watermarks, and to obtain such
marks by embedding a digital string into an image. The digital watermarks can be
created either in a spatial domain or in a transform domain. In transform domain
digital watermarking, the digital string is embedded using a transformation space,
such as Fourier transform or the like, to obtain a spread spectrum characteristic
of noise insensitivity. Correspondingly, the digital watermark is not an image and
can not be simply reconstructed or verified optically, though the actually encoded
digital watermark data can be arbitrary, thus also an image in some other transform
domain. An example of spatial domain digital watermarking is found in
U.S. Patent 6,252,971
to S. Wang, for "DIGITAL WATERMARKING USING PHASE-SHIFTED STOCLUSTIC SCREENS,"
issued June 26, 2001, where the watermark can be reconstructed as image directly
in the space domain.
Classical watermarks are image-wise watermarks, meaning
that they can be viewed by a user, normally to verify authenticity, but often only
to establish a perception of value or beauty. Image-wise digital watermarks exist
in the form of glossmarks, correlation marks and embedded image watermarks as described,
for example in
U.S. Patent Application No. 11/034,131
(above), as well as
U.S. Patent Application No. 10/876,001
U.S. Publ. 20050128524 A1
The advantage of such watermarks is that they more closely
resemble classical watermarks, albeit with the added capability of variable content.
This variable content, however, also directly interacts with the background image
and a truly automated use of variable watermarks requires the correct positioning
(or identification of the watermark location) based on the watermark content and
the image content. In other words, when an image-wise watermark is to be employed,
and it is desirable to either make the mark visible or to hide the mark, the "best"
or preferred position at which such a mark should be reproduced in the image is
strongly dependent upon the nature of the mark as well as the image - where simply
repositioning the mark within an image may significantly alter the detectability
(desired/undesired) of a mark.
U.S. Patent 6,956,958
describes a method of enhancing color fidelity in multi-reproduction that
includes an encoding part, which usually resides in a printer (or the printing part
of a copier, but may also be resident in software stored in a computer) and a decoding
part which usually resides in a scanner (or the scanning part of a copier, but may
also be resident in software stored in a computer) so as to add color fidelity information
directly to the output print. Each part can be divided into layers where the top
layers are responsible for management of color information while the bottom layers
are responsible for embedding/detection of digital watermarks containing the color
Accordingly, the present disclosure is directed to a system
and a method for determining preferred image locations for placing watermark information
for both hidden and obvious marks. Preferred locations provide either a high degree
of hiding or visibility depending on the goal. At least three metrics may be employed
in making a determination of a preferred position of a watermark, including (i)
image value, (ii) image/watermark value similarity, and (iii) image/watermark contrast
Disclosed in embodiments herein is a method for determining
a location for an image-wise watermark in relation to a target image, comprising:
analyzing the target image by dividing the target image into a plurality of smaller
areas and determining a characteristic metric for the areas thereof; and determining,
based upon the characteristic metric, at least one location for the watermark image
to be overlaid onto the target image.
Also disclosed in embodiments herein is a print production
system for determining a location for an image-wise watermark in relation to a target
image, comprising: a processor for analyzing the target image by dividing the target
image into a plurality of smaller areas and determining a characteristic metric
for the areas thereof; memory for recording the characteristic metric data for each
of the areas analyzed by the processor; said processor further determining, based
upon the characteristic metric, at least one location for a watermark image to be
overlaid onto the target image, and overlaying said watermark image on the target
image in the at least one location; and a printing engine for receiving the target
image with the overlaid watermark image and rendering the image on a substrate.
In a further embodiment the target image is a variable data image and where determining
a location for an image-wise watermark occurs on a document-by-document basis.
Further disclosed in embodiments herein is a method for
printing variable data with a watermark, comprising: receiving a target image into
which an image-wise watermark is to be embedded; receiving a watermark image; analyzing
the target image by dividing the target image into a plurality of smaller areas
and determining a characteristic metric for the areas thereof; analyzing the watermark
image by dividing the watermark image into a plurality of smaller areas and determining
a characteristic metric for the areas thereof; and determining, based upon at least
one of the characteristic metrics, at least one location for the watermark image
to be overlaid onto the target image.
BRIEF DESCRIPTION OF THE DRAWINGS
In a further embodiment the characteristic metric is image similarity between the
target image and the watermark image.
In a further embodiment the method further comprises determining image similarity
as a function of a variance between average gray levels for the areas of the target
image and the watermark image.
In a further embodiment the characteristic metric is image contrast.
In a further embodiment the method further comprises determining image contrast
as a function of a difference between contrast levels for the areas of the target
image and the watermark image.
In a further embodiment the characteristic metric is image value range.
In a further embodiment the method further comprises determining image value range
as a function of the range of image data within areas of the target image.
In a further embodiment the method further comprises receiving variable data for
the watermark image and generating a watermark image incorporating the variable
In a further embodiment the method further comprises receiving variable data for
the target image and generating a target image incorporating the variable data.
In a further embodiment the method further comprises determining the at least one
location for the watermark image as a function of a plurality of characteristic
FIG. 1 is a schematic illustration of a system and method
for carrying out an embodiment of the invention;
FIG. 2 is an illustrative example of a target image in
accordance with an aspect of the disclosed embodiments;
FIG. 3 is an illustrative example of the data derived from
an analysis of areas of FIG. 2;
FIG. 4 is an illustrative example of a watermark image
in accordance with an aspect of the disclosed embodiments;
FIG. 5 is an illustrative example of the data derived from
an analysis of areas of FIG. 4;
FIG. 6 is an illustrative example of the data array calculated
from a characteristic metric relating to the data of FIGS. 3 and 5 in accordance
with one embodiment;
FIG. 7 is an illustrative example of another target image
in accordance with an aspect of the disclosed embodiments;
FIG. 8 is an illustrative example of the data derived from
an analysis of areas of FIG. 7;
FIG. 9 is an exemplary output illustrating an aspect of
the disclosed embodiments; and
FIG. 10 is an alternative schematic illustration a variable
data system and method for carrying out an embodiment of the invention.
In the case of image-wise watermarking, two possible applications
/ intentions can be considered as the extreme cases, and all intermediate cases
can be derived from two end points established by the extreme cases. In the first
case, the watermark is intended to be as unobtrusive as possible, and in the second
case, the watermark is intended to be clearly visible. An example of the first scenario
is where the image-wise watermark is intended to be completely hidden. An example
of the second scenario is a variable data glossmark storing, for example a serial
number, on a photo.
The "best" area for embedding a watermark can generally
be found following a combination of one or more of three criteria: (i) image/watermark
value similarity; (ii) image/watermark contrast similarity; and (iii) image value
range. As will be appreciated based upon the following discussion, the use of the
criteria may be independent of one another, or two or more criteria may be combined
using a weighted average or similar approach so as to optimize the placement of
the watermark in order to achieve a desired objective of the watermark (e.g., maximize
or minimize visibility).
Generally speaking, with respect to the image/watermark
value similarity criteria, if the target image and the watermark image information
are similar, the watermark will be more subtle or less detectable. On the other
hand, if they are dissimilar, the watermark will be more obvious or apparent. As
for the image/watermark contrast similarity, if the contrast of the target image
is high, watermark contrast will be lost, and vice versa. Lastly, for the image
value range criteria, since all watermarks are created by modulating the input image,
sufficient image data must be present locally to serve as an information carrier
(to carry the overlaid watermark image).
Having described the general considerations used to select
locations for a watermark to be overlaid on a target image, attention is now directed
to a more detailed discussion relative to each of the possible criteria used to
determine a location. Referring to FIG. 1, in one embodiment the image value similarity
criteria computes a gray-level based variance (if the absolute brightness of the
watermark is not fixed, the calculation can be done bias independent). In order
to do this, the input or target image 110 is first received or retrieved
from a storage medium or memory 114 as indicated by operation 120.
An example of a target image is shown in FIG. 2 Next, the target image is divided
into or reduced to a small set of image areas 124, for example about 20 x
20 areas, as represented by operation 128. In the analysis phase of operation
128, the average gray level of the target image areas is computed and stored
as target data 132.
In a similar manner, the watermark image 140, having
been received and/or stored in memory, operation 144, is also reduced in
size to a substantially smaller number of areas, say 4 x 4, as depicted at
154. The divided watermark image is characterized as indicated by reference
numeral 150, and the associated characteristic metric data indicated as
158. An example of a watermark image is depicted in FIG. 4
Processing of the images in this manner can be justified
by the application, where the watermark only covers a certain area of the image,
or is replicated at several points in a target image, with one of the replications
being in an optimized spot, or by identifying important and not so important areas
of the watermark. In general, it can be said that if a position can be chosen, it
also means that the watermark must be smaller than the target image; otherwise,
no choice would be available. Moreover, while the disclosed embodiment is directed
to particular images and relative number of areas or sections it should be recognized
that these are illustrative examples, and that the systems and methods described
relative to the examples have broad application and are not necessarily limited
by the example images or the exemplary analysis details set forth.
FIG. 2 shows an example target image and FIG. 3 the corresponding
area entries indicating the average intensity values for each of the 20 x 20 image
areas. The array 132 of average intensity data depicted in FIG. 5 is an example
of the data that is described above as being derived from the target image. It should
be noted that the number of areas (e.g., 20 x 20) is chosen somewhat arbitrarily
and that equivalent calculations can also be performed on various array sizes. It
is also the case that equivalent calculations can be performed for color images
and color image data by, for example, looking at the luminance for scalar data or
by looking at the color for vector computations.
In order to embed or overlay the watermark image shown
in FIG. 4, it is necessary to compute the corresponding image areas of the watermark
image as illustrated in the array of FIG. 5. After calculating the variance values
between the target image and the watermark, the resulting variance being depicted
in array 610 of FIG. 6, it is then possible to analyze or sort the variances,
rendering - in this case - the five highlighted entries (620) in the array
of FIG. 6 as the lowest variance areas, and thus the optimized locations for embedding
of the watermark.
Returning to FIG. 1, using only the similarity criteria
for example, the method of determining an optimal location of the watermark is completed
by the comparison described above relative to the examples, where the resultant
array of FIG. 6 (divided image areas 620) are identified at operation
170, and the watermark is then overlaid or embedded into the target image
at those locations, producing the watermarked digital image 174.
A similar processing can be performed for the contrast
criteria at operations 128 and 170 if FIG. 1. A second embodiment
of the method is to use an automatic image enhancement (AIE) based contrast calculation.
Such a characteristic metric may be established by following the method outlined
U.S. Patent 5,450,502
for a "IMAGE DEPENDENT LUMINANCE ENHANCEMENT," by R. Eschbach et al. issued
September 12, 1995, and hereby incorporated by reference in its entirety. For certain
kinds of watermarks, it can be assumed that the watermark is best located in image
regions of low contrast. This can be done by separating the image into a number
of regions. For simplicity, assume non-overlapping regions, though overlapping and
multi-scale regions are understood variations of the described approach. Assuming
10 x 10 areas as used for example in Figure 7, it is possible to compute the local
contrast, preferably using the histogram variance method outlined in
U.S. Patent 5,450,502
. In the histogram variance method, N histograms are derived - in the example
10 x 10 = 100 histograms - that are then sorted according to variance of the histogram,
with a high variance indicating a low contrast and vice versa. One can now use a
fixed threshold to find areas of a prespecified low contrast attribute, or one can
alternatively find the M areas with the lowest contrast (largest variance value)..
FIG. 7 shows a target image and the array 810 in
FIG. 8 gives the corresponding overlay areas for 10 x 10 areas of the target image,
assuming a fixed threshold in the contrast variance metric. In the disclosed example,
it is assumed that the "best" overlay areas are those with the least image contrast
(indicated by an "X" in array 810), since the image contrast will act as
localized noise with respect to the watermark. It will, however, be appreciated
that this applies in the case where it is desirable to view the watermark - for
example a visible watermark overlay using a glossmark. In the opposite case, where
desirable to hide the watermark, it may be preferable to place the watermark in
areas having greater image contrast.
Yet a third embodiment for the method by which an optimal
watermark location is determined involves the image value range characteristic as
noted above. In this embodiment, perhaps the simplest, area selection is based on
the input image value range. The basis for this characteristic is the assumption
that any coding or modulation of the image data to embed a watermark in the target
image data requires that the target image act as an information carrier. Hence,
the data of the target image must provide sufficient image data in a local area
to serve as the information carrier for the watermark. The image value range is
simply defined based on the watermark criteria, and in general will limit the useful
area to exclude the extreme ends of the image. In other words, areas of the target
image that would be most suitable for coding a watermark are mid-tone areas, whereas
the totally white areas or fully saturated color areas would be the least suitable.
The following example illustrates the importance of the
system and methods disclosed herein to automatically identify preferred or "best"
watermark locations in a target digital image. Consider the scenario, in which a
unique ID number is to be placed visibly over an image. In order to understand this
example the reader is referred to FIG. 9, where there is depicted an identification
card 810 having a person's photograph in region 820 overlaid with
a repeating ID number 830. Although for purposes of illustration the ID number
was simply placed over the image, in actual applications the watermark (ID number)
would typically be much less visible. In the example, the ID number 830 is
repeated over the image 820. On careful examination, one sees that the digits
in several regions (e.g., upper right region 850) are virtually invisible
in the image, despite the extreme nature of this type of embedding. When using any
form of watermark, the general watermark contrast will be reduced and the numbers
will be much less visible in order to not to destroy the image for identification
Knowledge of the watermark size, as well as knowledge of
the optimized watermark areas would allow an automated system to assure that the
complete ID is readable, even though parts of the ID are obscured in every instance
of the ID number string (e.g.: the first portion of the ID is visible in one region
of the image, a second portion in another region, and a last portion in yet a further
region). This will encapsulate sufficient overlap to uniquely read the entire ID
number even though there might not be a single area in the target image sufficiently
large to hold the entire watermarked ID number.
As briefly mentioned above, the three criteria described
in detail above may be employed, alone or in combination with one another (or other
criteria) to determine a preferred location within an image to apply a watermark.
Use of the system and method described above also requires that an instruction or
requirement be established to indicate whether the process is to optimize visibility
or "invisibility" of the watermark (i.e. overlaying in dissimilar or similar image
areas). Next, a determination must be made as to which of the criteria will be employed
to make the location determination. If more than one criterion is employed, the
weighting of each criterion should also be established in relation to the particular
intent of the watermark. Such weighting or other parameters (e.g., thresholds for
contrast, variance, etc.) are then employed during the processing of the target
image and/or the watermark image. Both, as indicated in FIG. 1 are processed to
produce a "low resolution map" of the image, such as the 20 x 20 and 4 x 4 arrays
for the target (132) and watermark (158) images respectively. Subsequently,
the analysis of the images proceeds in accordance with the selected criteria ((i)
image/watermark value similarity, (ii) image/watermark contrast similarity; and
(iii) image value range), and the result is likely an array 610 as depicted
in FIG. 6, where the processed data is retained as a "map." The areas of the map
can then be sorted (based upon the predefined preferences (e.g., ascending or descending
order), and the first N areas will then be identified as the preferred locations
into which the watermark is embedded.
As a further example, consider the situation where a glossmark
is to be embedded into a target image so that upon printing of the target with the
glossmark, the mark will be visually unobtrusive, but nonetheless detectable. In
such a situation, the image value range criteria is of particular import because
areas with too much or too little toner are undesirable for the embedding of glossmarks.
Hence, the image value range criteria would be used to rule out certain areas of
the target image. Next, the contrast similarity criteria would be applied, but in
this case to the target image only; assuming that even low contrast portions of
the glossmark should be visible results in a "constant" contrast map for the glossmark
image. The result of the contrast similarity criteria would then be employed to
select areas with minimal variance as preferred locations for embedding the mark.
In yet a further application of the methods described herein,
the areas identified for embedding the mark may be further employed to control the
size of the watermark to be applied. For example, if a plurality of adjacent areas
are suitable (e.g., above a contrast variance threshold), then it may also be possible
to agglomerate the areas so as to provide a larger region (consisting of adjacent
analysis areas) in which the watermark may be embedded. In other words, the size
(magnification) of the watermark may be increased to cover more than a single area.
The disclosed system and method further find application
in the production of variable or on-demand documents as depicted for example in
FIG. 10, particularly when such documents need to be rendered with an image-wise
watermark. Recognizing that in variable print output (variable input target images
1010), the target image may vary with each printed document. In such a situation,
the previously described process may be applied on a document-by-document basis
in order to assure that the resulting output documents 1074 each have a watermark
displayed in a preferred location of the output document - regardless of the fact
that the content of the document may change from one document to the next. It will
be appreciated that the system and methods described above emulate, to some level,
what experienced graphic designers might do in selecting the location(s) for watermarks
- thereby adding "intelligence" as opposed to simply applying a watermark redundantly
over entire image.
Those knowledgeable with variable data printing system
will recognize that the schematic illustration of FIG. 10 identifies several components
and nodes of know production printing environments. The various operations described
herein may, therefore, be performed on a digital front end (DFE) device
1080 or other prepress processors that are suitable for receiving image data,
including watermark images and variable input data, storing the data I a memory
and then processing the data to produce a stream of information that may be employed
to drive a printing system (printer 1090) such as a digital xerographic printer
- whether black/white, highlight color or full color. The output of such printers
may, therefore, incorporate variable print information as well as optimally placed
image-wise watermarks, without the requirement of user intervention.