1. Field of the Invention
This invention relates to systems and methods for determining a location of a visual link to digital media on physical documents, and more particularly to calculating and adjusting a size and location of a visual digital media link based on the location of other digital media links and the similarities between portions of content on the media.
2. Description of the Related Art
There are many systems that provide links to digital media from physical media—such as a paper document—using visual content analysis and indicators. Typically, a user takes a picture of the document that includes a visual link using a cell phone or other portable device with a camera, and the image is processed either on the phone or a remote server to identify and then access the linked media. The linked media is then displayed on the phone.
One example of a visual link is an Embedded Media Marker (EMM) 100, illustrated in
The existing EMM system includes an authoring tool for placing an EMM on a document page that allows the user to interactively select a region of the page that is meaningful for EMM placement, and that has a number of keypoints for recognition. A keypoint is a feature vector describing the local image characteristics near a location in an image. As the user moves the EMM to desired locations, the capture region is dynamically increased or decreased based on the number of document keypoints. If there are insufficient keypoints, the EMM signified region of the page is shaded red. Although this prevents the user from placing an EMM in a region with too few keypoints, it does not guarantee recognition accuracy, since the keypoints might be similar or identical to other EMMs in the database.
While the current authoring tool in the EMM system checks if there are enough keypoints in a candidate region for an EMM, it does not check if those keypoints would conflict with previously authored EMMs or if the keypoints could be easily confused with content on other pages in a collection.
Systems and methods described herein provide for determining a location and size of a visual link to digital media on physical media, and more specifically to an authoring tool for creating Embedded Media Markers (EMMs) on one or more pages of a physical document, identifying and scoring similarities between the newly-created EMM and other EMMs and keypoints on the physical document, and providing a visualization of the scores to a user who is creating the EMMs so that a newly-created EMM can be positioned and sized on the physical document to avoid confusion with other EMMs and related content.
In one embodiment of the invention, a method for placing a visual link to digital media on a document comprises displaying at least one page of a document on a display; identifying and matching keypoints of the document with existing visual links and corresponding keypoints using a computer with a processor and a memory; scoring keypoints based on similarities between the keypoints of the document and keypoints of the existing links; receiving a user input selecting one or more regions of the document for placement of the visual link; and placing the visual link on the document.
The score of each keypoint may be modified based on scores of surrounding keypoints.
A visual overlay may be generated on the document based on the keypoint scores, wherein the visual overlay provides a visual indication of the suitability of one or more regions of the document for placement of the visual link.
The size and location of the visual link may be adjusted based on the keypoint scores.
The keypoints adjacent to a keypoint which all match an existing visual link may be penalized.
The modification of the score of each keypoint may be further modified by a weighted scale corresponding to a distance between the keypoints.
The suitability of a region of the document for placement of the visual link may be determined based on a keypoint score density.
The visual overlay may display gradients of shading to indicate whether one or more regions are suitable for placement of the visual link.
The suitability of a region of a document may be increased where a plurality of keypoints within a set radius match a visual link.
The visual links may be embedded media markers (EMMs).
In another embodiment of the invention, a system for placing a visual link to digital media on a document, comprises a display unit which displays at least one page of a document; a matching unit which identifies and matches keypoints of the document with existing visual links and corresponding keypoints; a scoring unit which scores keypoints based on similarities between the keypoints of the document and keypoints of the existing links; an input unit which receives a user input selecting one or more regions of the document for placement of the visual link; and a placement unit which places the visual link on the document.
A score modifying unit may modify the score of each keypoint based on scores of surrounding keypoints.
A visualization unit may generate a visual overlay on the document based on the keypoint scores, wherein the visual overlay provides a visual indication of the suitability of one or more regions of the document for placement of the visual link.
The size and location of the visual link may be adjusted based on the keypoint scores.
The keypoints adjacent to a keypoint which all match an existing visual link may be penalized.
The modification of the score of each keypoint may be further modified by a weighted scale corresponding to a distance between the keypoints.
The suitability of a region of the document for placement of the visual link may be determined based on a keypoint score density.
The visual overlay may display gradients of shading to indicate whether one or more regions are suitable for placement of the visual link.
The suitability of a region of a document may be increased where a plurality of keypoints within a set radius match a visual link.
The visual links may be embedded media markers (EMMs).
In yet another embodiment of the invention, a computer program product for placing a visual link to digital media on a document may be embodied on a computer-readable medium and when executed by a computer, performs the method comprising displaying at least one page of a document on a display; identifying and matching keypoints of the document with existing visual links and corresponding keypoints using a computer with a processor and a memory; scoring keypoints based on similarities between the keypoints of the document and keypoints of the existing links; modifying the score of each keypoint based on scores of surrounding keypoints; generating a visual overlay on the document based on the modified keypoint scores, wherein the visual overlay provides a visual indication of the suitability of one or more regions of the document for placement of the visual link; receiving a user input selecting one or more regions of the document for placement of the visual link; and placing the visual link on the document.
Additional aspects related to the invention will be set forth in part in the description which follows, and in part will be apparent from the description, or may be learned by practice of the invention. Aspects of the invention may be realized and attained by means of the elements and combinations of various elements and aspects particularly pointed out in the following detailed description and the appended claims.
It is to be understood that both the foregoing and the following descriptions are exemplary and explanatory only and are not intended to limit the claimed invention or application thereof in any manner whatsoever.
The accompanying drawings, which are incorporated in and constitute a part of this specification, exemplify the embodiments of the present invention and, together with the description, serve to explain and illustrate principles of the invention. Specifically:
In the following detailed description, reference will be made to the accompanying drawings. The aforementioned accompanying drawings show by way of illustration and not by way of limitation, specific embodiments and implementations consistent with principles of the present invention.
Systems and methods provide for determining a location and size of a visual link to digital media on physical media such as a paper document. An authoring tool for creating a link on a document, such as an Embedded Media Marker (EMM), identifies and scores other EMMs and related keypoints on the document to determine similarities between a newly-created EMM and other EMMs and keypoints on the paper document. The scores are visualized for a user on a display in order to position and size the newly-created EMM in a location on the paper document that will avoid confusion with other EMMs and related content. The location and size of the newly-created EMM may be automatically adjusted based on the scoring of the keypoints and related EMMs. The system for authoring EMMs on a collection of document pages allows each of the EMMs to be accurately identified.
The EMM system places a mark on the paper document indicating a specific region of the page with a link to media. Described herein is a system for authoring EMMs on a collection of document pages such that each of the EMMs can be accurately identified when captured by a user using a portable device. For each keypoint in a candidate page region, approximate neighbors are located. Keypoints with close neighbors are assigned negative scores that are spread in the vicinities of those keypoints. A visualization technique indicates page regions with predominantly negative scores so that the author can avoid them. The system also automatically resizes and moves EMMs being authored to avoid such regions.
The embodiments described herein also introduce an EMM authoring tool that takes into consideration possible matches of regions on a page both against previously authored EMMs and against all other pages in a collection. Close matches indicate potential for confusion in later retrieval attempts. Thus, the authoring tool steers the user away from problematic regions on a page and towards regions that have a large number of unique keypoints.
I. Overview
An overview of a method for placing an EMM on the physical document will now be described and is correspondingly illustrated in
Different approaches for keypoint detection and description of their features can be used, such as that described in U.S. application Ser. No. 12/646,841, filed Dec. 23, 2009 and incorporated herein by reference in its entirety. Similarly, different database representations for locating nearby keypoints can be used. An approximate nearest neighbor (ANN) tree would work well for this purpose. The only requirement is that keypoints can be detected in an image, that nearby keypoints can be found in a database, and that the found keypoints include distances to the image keypoints and associations to previously authored EMMs.
II. Collections of Document Pages
An authored collection of EMMs comes from a corresponding collection of document pages. Such a collection may consist of the pages of a single book or issue of a magazine. A collection may also encompass the pages of all the issues of magazine published during a time period. In the latter situation, the collection may change over time when new issues of the magazine are added. Readers of such books, magazines, or other documents start by identifying the collection. They then take a picture of an EMM using a portable device with a camera—such as a smartphone, tablet, or personal digital assistant (PDA)—that uploads the picture to an EMM server and redirects the device to the media link associated with the EMM.
The authoring tool presents the pages of the collection as image thumbnails to allow the author to select a page and to place one or more EMMs on it. Especially in cases where the collection does not grow over time, it is beneficial to compare the page shown in the authoring tool to the other pages in the collection and to indicate page regions that can be confused with regions on other pages.
III. Improving Retrieval Performance
When trying to determine a matching EMM after taking a picture of a part of a document page, the system first determines the rotation- and scale-invariant keypoints in the submitted picture. For each keypoint, the closest or approximate closest keypoint in a database of all keypoints of all EMMs is determined. The number of found keypoints is accumulated for each EMM and the EMM with the most matching keypoints is returned, provided that the keypoint count exceeds a threshold.
According to one embodiment, the authoring tool improves the retrieval performance by avoiding page regions that would produce ambiguous matches. Keypoints from candidate page regions are checked against a keypoint database. If there are too many matches, the author is discouraged from using a particular candidate region. In one embodiment, the author may be discouraged from using a particular region by three visual cues: the overlay will turn a dark brown color, the size of the EMM may become larger when moving it to the region, and the circle turns red if there is no possible location to place the EMM in that region. The size and color of the EMM may change instantly and in real-time as the user moves the EMM around the page to find a suitable location. Alternative visual or audio prompts could be given, such as changing a cursor shape when a bad region is located or using an area of the interface to display a message to the user indicating whether a region is suitable for placement or not.
IV. Looking at Other Pages in the Collection
When an author works on a page to place one or more EMMs, all the keypoints on that page are matched against a database of existing keypoints. Those keypoints come either from previously authored EMMs or from other pages in the same collection. Other pages are represented as one or more EMM. To support authoring of EMMs, it is advantageous to segment the page using either a simple grid (e.g., a 3×4 grid for a typical portrait page) or the results from a page segmentation algorithm. When a page is segmented, each segment is treated as an EMM. Because the algorithm is the same in both cases, we refer to the other pages and page segments as EMMs for the rest of this section. Comparing the EMM authoring page against the other pages in the collection has both advantages and disadvantages. An advantage is that regions that are ambiguous such as text passages in the default font can be avoided. The same characters or words appear on many different pages so that it is possible that another EMM with those characters or words would provide a better match to a low-quality picture of an EMM containing such an ambiguous page region. Such regions can be found without requiring the authoring of previous EMMs. On the other hand, such an approach (if used to block the authoring of EMMs) may prevent attaching an EMM to a distinctive region that is repeated on two or more pages.
For authoring EMMs, when a user indicates the desire to author an EMM for a region that is repeated some limited number of times, the system can ask if the EMM should be added for all instances of this region. This would make it possible to map all instances of a corporate logo to the same destination and make it clear to the author that visually similar EMMs must link to the same location.
V. Determining Possible Conflicts During Authoring
To determine which keypoints on the EMM authoring page conflict with other pages or EMMs, each keypoint is assigned a score between −1 and 1. Keypoints that do not have a match to a keypoint in the database within the distance threshold receive a score of 1, indicating that there is no conflict. Other keypoints are assigned a score s based on the distance d and the maximum match distance dmax. The normalized distance between 0 and 1 is linearly mapped to a score between −1 and 1.
s=2*d/dmax−1 (1)
When evaluating a region on the authoring page, the scores of keypoints in that region are added. A large positive sum indicates that the region is suitable for placing an EMM. A neutral or, even worse, a negative sum indicates that the region is unsuitable. A threshold may be used to determine if a page region contains enough keypoints. For example, the page region could be required to contain at least 100 keypoints. By using the sum of keypoint scores instead of the keypoint count, the notion of conflicts can be incorporated with other pages or EMMs. For example, either 100 keypoints with a score of 1 or 200 keypoints with an average score of 0.5 would be sufficient for a page region. To improve performance for finding suitable page regions, only rectangular regions parallel to the page bounds are used. However, this approach can also be applied to non-rectangular regions, for example, by first looking at the rectangular bounding box of the region and then checking the actual shape in a second step.
To further speed up the computation of sums of keypoint scores, the use of an integral map is adapted that pre-computes the cumulative score for each point on the page from the page origin, as disclosed in U.S. Ser. No. 12/646,841 (referenced above). For determining the sum of keypoint scores in a rectangular region parallel to the page bounds, only four entries in the integral map have to be looked up and added or subtracted.
VI. Emphasizing Conflicts with the Same EMM
Because keypoints that match the same EMM would bias the retrieval towards that EMM at the expense of the newly authored EMM, the scores for those keypoints are lowered. This difference between the perfect score of 1 and the actual score of a keypoint is called the penalty score. As the keypoint score lies between −1 and 1, that penalty score ranges from 0 to 2. The system spreads this penalty to nearby keypoints that match the same EMM. The penalty score is weighted by a distance function and the weighted sum of matching penalty scores is accumulated at each keypoint. This weighted sum of penalty scores is subtracted from the original score of the keypoint.
The two keypoints matching E3 are further apart and hypothetically share 10% of their penalty score with each other. Thus, if they had the same initial scores as the E1 pair (0.0 and −0.5), their final scores are −0.15 (=0-0.1*(1-−0.5)) and −0.6 (=−0.5-0.1*(1-0) respectively. The top three keypoints (matching E2) provide an example where the middle point would be reduced by both of the other points while the far left and far right point are beyond the threshold for penalty sharing.
By penalizing nearby keypoints matching the same EMM further, page regions containing these keypoints are made even less suitable. Because those regions would easily be confused with the matching EMM, this is a desired outcome.
In one embodiment, the spread penalty score is weighted by a normalized distance within a given radius. A radius of 0.25 inches may be used, but other radii may produce better results with different source materials. Rather than reducing the spread penalty score linearly with distance, other mathematical functions such as the square distance, a Gaussian function, or a constant factor with a distance cutoff are also possible. The spread penalty score is accumulated separately and added to each keypoint score in a separate step to avoid spreading beyond the distance threshold or spreading back and forth. The combined score can be truncated at −1 to avoid large negative values. However, the visualization described in the next section works better without this truncation in dense areas of keypoints matching the same EMM. As shown in
When determining which keypoints match EMMs in the database, it may be opportune to look at the k closest matches rather than just looking at the closest match. Such an approach deals with situations where many keypoints in a region match the same EMM but some of those keypoints have slightly closer matches for different EMMs. This approach prevents that a keypoint of a conflicting EMM is shadowed by a keypoint from another EMM. Such a region would still be unsuitable for placing a new EMM. Thus, the penalty score should reflect this unsuitability. Rather than just spreading one penalty score, penalty scores for different matching EMMs can be spread in parallel. For each keypoint, the penalty score for each match may be separately computed. Those penalty scores are then spread separately such that each keypoint in parallel accumulates weighted penalty scores for each EMM in the database that it matches. The maximum of the weighted sums of penalty scores is subtracted from the original score.
As an alternative, one can use a different score for each match based on weights. The weight wi for a match is based on the distance of that match, di, and the distance of the next closest match, di+1.
wi=di+1/di (2)
Given this weight, the score for the match would be expressed as:
si=2*(1−wi)−1 (3)
A disadvantage of this variant is that worse matches can be assigned higher weights. For example, if the two best matches have similar distances and the third-best match is much worse, the second-best match would be assigned a higher weight than the best match.
VII. Using Nearby Matches During Retrieval
When taking multiple matches into consideration during retrieval, the plain voting scheme of counting the number of keypoint matches for each candidate EMM is replaced by sums of keypoint match scores for each candidate EMM. A keypoint match score is 1 at a distance of 0 and 0 at the maximum match distance. A variant of the penalty score described in the previous section can be used to boost the score for candidate EMMs where several keypoints within a radius match the same EMM. The penalty score is highest when the distance to a matching keypoint is smallest. The same is desirable for a bonus score. The keypoint match score can serve as the bonus score without modification. Just like the penalty score, the bonus score is spread within a radius to keypoints matching the same EMM. Scores are accumulated in parallel for different candidate EMMs. Unlike the penalty score where only the maximum accumulated score is used, all accumulated bonus scores are taken into consideration when determining which candidate EMM has the highest score total.
VIII. Visualizing Page Regions Suitable for Authoring EMMs
As illustrated in
Regions suitable for EMMs contain many keypoints not similar to those used by other EMMs and few keypoints that are already used. When just looking at the presence of keypoints, the average keypoint density in a region indicates its suitability. An application-specific threshold for this density can be used to visualize if the density is above the threshold or how close it gets. In one embodiment, a threshold of 100 keypoints in a 2-inch by 2-inch region may be used.
To determine the average keypoint density while taking keypoint scores into consideration, the score of each keypoint is spread over a region. The score is spread over a circular region with a weight that linearly decreased with distance. A radius of 0.5 inches is used for the circular region, but other radii may produce better results with different source material. Other functions can be used as well, such as a two-dimensional Gaussian function or a square region where the weight either decreases with distance or stays constant. In all cases, the integral under the two-dimensional curve should be normalized to one.
This approach can be used with the negative keypoint scores described earlier. When combining positive and negative scores, both an empty region and a region with an equal amount of positive and negative scores would have a neutral suitability. When all keypoints have a score of 1, any region where the keypoint density exceeds the threshold has the maximum suitability. A higher density of keypoints with positive scores less than one would also lead to the maximum suitability. Conversely, a high density of keypoints with negative scores produces a region with a negative suitability.
When spreading the earlier discussed penalty score that penalizes keypoints in the same vicinity matching the same EMM, keypoint scores can drop below −1. When combined with the keypoint score density, this produces desirable results because those scores push a larger region to the minimum suitability, indicating to the user that this region should really be avoided.
IX. Automatic Adjustment of EMM Location and Size
Using the sums of keypoint scores described above, the authoring tool automatically adjusts the size of the EMM as the user places it on the page or drags it to a different location. The authoring tool determines the smallest rectangular shape at the same aspect ratio and centered on the location specified by the author that has a sufficiently large total score. If the resulting regions would be very large, the authoring tool attempts to move the EMM to avoid unsuitable regions. Both assistive techniques can be turned off by the author. If no suitable region can be found, the EMM is colored red to indicate its unsuitability.
X. Computer Embodiment
The embodiments and implementations described above are presented in sufficient detail to enable those skilled in the art to practice the invention, and it is to be understood that other implementations may be utilized and that structural changes and/or substitutions of various elements may be made without departing from the scope and spirit of present invention. The following detailed description is, therefore, not to be construed in a limited sense. Additionally, the various embodiments of the invention as described may be implemented in the form of software running on a general purpose computer, in the form of a specialized hardware, or combination of software and hardware.
Number | Name | Date | Kind |
---|---|---|---|
6711293 | Lowe | Mar 2004 | B1 |
20100080469 | Liu et al. | Apr 2010 | A1 |
20100259537 | Ben-Himane et al. | Oct 2010 | A1 |
20110072037 | Lotzer | Mar 2011 | A1 |
20110154174 | Liu et al. | Jun 2011 | A1 |
20110239296 | Schipper | Sep 2011 | A1 |
20120037695 | Liu et al. | Feb 2012 | A1 |
20120177002 | Faucher et al. | Jul 2012 | A1 |
20120239647 | Savenok et al. | Sep 2012 | A1 |
20120269439 | Yang et al. | Oct 2012 | A1 |
Entry |
---|
Liu et al. “Embedded Media Markers: Marks on Paper that Signify Associated Media”. Copyright Date: Apr. 9, 2009. |
Liu et al. “Embedded Media Barcode Links: Optimally Blended Marcode Overlay on Paper for Linking to Associated Media”. Copyright Date: 2010. |
Li et al. “Ways to Integrate Digital and Physical Media by Using Middleware.” <no Copyright Date>. |
Yang et al.“Large Scale EMM Idetification Based on Geometry—Constrained Visual Word Corresponding Voting.” Copyright Date: Apr. 20, 2011. |
Yang et al. “Minimum Correspondence Sets for Improving Large-Scale Augmented Paper.” Copyright Date: 2011. |
Girgensohn et al. “A Tool for Authoring Unambiguous Links from Printed Content to Digital Media”. Copyright Date: 2011. |
Sunil Arya, et al., “An Optimal Algorithm for Approximate Nearest Neighbor Searching in Fixed Dimensions,” The Proceedings of the Fifth Annual ACM-SIAM Symposium on Discrete Algorithms, 1994, pp. 573-582. |
Qiong Liu, et al., “High Accuracy and Language Independent Document Retrieval With a Fast Invariant Transform,” IEEE International Conference on Multimedia and Expo (ICME), 2009, 4 pages total. |
David G. Lowe, “Distinctive Image Features from Scale-Invariant Keypoints,” International Journal of Computer Vision, Jan. 5, 2004, pp. 1-28. |
Xu Wangming, et al., “Application of Image SIFT Features to the Context of CBIR,” 2008 International Conference on Computer Science and Software Engineering, 2008, pp. 552-555. |
Number | Date | Country | |
---|---|---|---|
20130036345 A1 | Feb 2013 | US |