The reason for the problem is that most search tools, including image search tools, don’t actually look at image, they look at the text around it. So unless the title of your work or the file name remain the same, there isn’t much hope for spotting a duplicate via traditional means.

However, in 2008, Tineye changed the game by introducing an image-based image search engine that would fingerprint images found online and match them against files uploaded to their service.

But, while Tineye’s matching technology is and always has been great, it’s been limited by Tineye’s rather small database. While that database has grown over 2x since I first wrote about the service (currently at just over 2 billion images), it hasn’t kept pace with the images being uploaded to the Web (Photobucket alone has 9.5 billion images).

However, last year Google entered into the fray, adding the ability to search for images by uploading or linking to another one, as with Tineye. The feature, which was initially an extension of Google’s previous Similar Image Search function, wasn’t very successful at first. However, over time, it appears Google has gone a long way to improving the tool as, in a recent spate of tests, it drastically outperformed Tineye in finding matching images consistently.

So, for artists looking to find their images on the Web, there seems to be a new king in town and it’s the same one authors have been using for years.

How to Use Google Search By Image

The easiest way to use Google Search By Image is to install either the Firefox
or Chrome extension, which simply enables you to right click an image to search for it.

However, barring that, you can simply visit Google Image Search and click the camera icon in the search bar.

That will open up the window that prompts you to either upload an image from your computer or provide the URL for one already online.

After you submit your image, Google will present a set of results. However, rather than being a “grid” like a regular Google Image Search, the results are ordred, first by exact matches and then by similar ones.

For example, I uploaded my recent image of the cover of Ocean’s Donkey Kong unlicensed port (from my recent article on video game plagiarism) and Google returned not only my article but other images containing that exact image.

In some cases, if there are multiple versions of the images but at different sizes, Google may suggest you look for alternate sizes of the image, as it did with the Limbo of the Lost cover from the same article.

If so it’s worth clicking the link to get a good breakdown of the other sizes (and places) the image appears. However, the real results, especially for images that have been widely copied, are below and you can see them by going through the various pages, as you would with a regular Google search.

But while Google Search By Image is cetainly easy to use, how well does it stack up against Tineye? The answer, is very well.

Comparison Tests

To test the two services head-to-head, I decided to have them both look for five different images used in recent articles on Plagiarism Today. These images are all either freely-available stock photos or are widely-used cover art for video games or records.

With that in mind, here’s the results of the tests:

Test 1: Angry Farm Image

Tineye Results: 1
Google Results: 558 (about)

Test 2: Generic Chart Image

Tineye Results: 4
Google Results: 555 (about)

Test 3: Facepalm Image

Tineye Results: 21
Google Results: 850 (about)

Test 4: Skull on Grave Image

Tineye Results: 1
Google Results: 3 (Not counting matching “Very Similar” results)

Note: Google’s “Very Similar” results were useless in this case as it just found other black and white photos without much regard for things that looked like the original.

Test 5: Bing Crosby White Christmas

Tineye Results: 173
Google Results: 883 (about)

Please note that the greatest limitation of this test is that I had to rely on both search engines to self-report how many matching images they had. However, I checked several pages of results with each test to make sure that the results were as accurate as possible.

Clearly though, the winner is Google, which found, in many cases, over 100x more matching images than Tineye. My suspicion is that, while Tineye’s matching algorithm is better (much fewer false positives), Google’s large database simply makes up the ground and then some, making it a much more valuable tool for image detection.

However, this doesn’t mean that Google’s perfect, there are still a few concerns and problems I have with it.

Limitations of Google

The biggest problem with Google is that, currently, there is no way to do a bulk search for a lot of images nor is there a way to do a recurring search. Though Google has an API for its Google Image Search, it doesn’t appear to work with Search by Image. Likewise, Google Alerts doesn’t allow you to create an alert for a Google Image Search at all, text or by image.

In short, there are no tools to make such searches easier and there aren’t likely to be any in the near future.

Meanwhile, Tineye has a very robust and well-established API that enables toe construction of just such tools.

Still, given how simple it is to use Google Search By Image, even with having to do the searches by hand, it’s still faster and easier than most methods, it’s still free and, in the end, just more effective, even more so than many paid-for tools.

There’s little reason not to integrate Google Search By Image into your checks and to use it at least some in your searches.

Bottom Line

When Google first launched this product, I tested it and found it to be less-than-useful. The algorithm was too flawed (based on the earlier and more limited “similar” search feature) and the number of false positives simply too high. Clearly, Google has made some great strides in the last six months and pushed this service to the point where it’s database and accuracy combine to make it the most useful image search tool available to the public, especially for free.

In the end, while I like Tineye as a company and as an offering, their database is too small and too limited to compete right now. Hopefully though, this competition will motivate both companies to improve their offerings and create a set of solutions that make things much easier on photographers and artists everywhere.

When people talk about content detection, they are usually putting it into some copyright-related context. Whether it is YouTube’s content ID system, image matching for tracking plagiarists/preventing orphans or simple duplicate text searching to track violators of their license, most people think of content detection as a means to track and stop copyright violations.

Sadly, this site too is guilty of that. However, I want to take time today to highlight one of the more important uses of content detection, audience analytics.

Most bloggers put some kind of analytics on their site to track visitors, referrals, etc. But if their revenue doesn’t come from online ads, it is more about understanding your audience than it is tracking actual page views. The statistics themselves are just a means to an end.

However, there is a very good chance that a significant portion of your audience is actually on other sites. But unless you track your content, you may never be aware of it.

Every Site is Different

Last year content tracking service Attributor announced the results of its TrueAudience study. It found, for the publishers that it checked, that the off-site audience was 1 and a half times greater than the audience on the site itself. This meant that, for every two people reading the content on the publisher’s site, three were seeing it elsewhere.

Since the Attributor study focused more on larger publishers, who will likely have higher levels of copying, the results will obviously not be that dramatic for smaller bloggers. However, virtually everyone who publishes to the Web will see some copying and, through that, will have some of their audience on other sites.

Ignoring this is like telling your statistics program to ignore every X visitor without any understanding of what number X is. If you want to know your audience, you have to go where they are.

Given the wide range of sites and the different situations they are in, it is impossible to even offer good estimates without at least getting some facts.

Referrals and Linkbacks

To be certain, you can track some of this with your existing tools. Referrals will alert you to when someone visits your site from another page and trackbacks/pingbacks will alert you instantly when someone has linked to your content.

However, there are several problems with these. Referrals are limited to when people actually click links. This requires both the site to link to you and a user to actually click the link. Given that the vast majority of visitors don’t click referral links, it could never actually happen, even with attributed uses. Also, referrals also track sites that simply link without using any content, making it a challenge to find actual audience members on other sites.

Though trackbacks and pingbacks don’t require anyone to click the link, they also focus mostly on sites that simply link to your blog. Furthermore, there is a huge issue with spam and many sites that duplicate your content may be filtered out, correctly or incorrectly, as such.

These tools are powerful, but they are not actual substitutes for following your content on the Web.

Getting Started

If you’re new to the idea of tracking your content, meaning you probably aren’t a regular reader of this site, here are a few suggestions to get you started.

1. FairShare: A free service provided by content tracking company Attributor, FairShare subscribes to your RSS feed and publishes a private one for you that tracks where it finds your content. Very useful for sites with a low-to-moderate level of copying.
2. Tineye: Though somewhat limited, Tineye is the best visual search engine available and definitely the best free search. Great for visual artists to find how their work is being used.
3. Plagium: If you have static content and can’t use FairShare, Plagium is a good alternative. Receive weekly alerts of new matches for free, works like a hybrid between Google Alerts and free Copyscape, other tools well worth looking into.

These are all great, free services that you can use to track your content and get a slightly better understanding of your audience.

Bottom Line

When I was running a personal literature site, I was proud of my traffic stats but was stunned to find out that my audience off the site was many times larger than it was on my site. Much of the use was legitimate, including use in compilations and online magazines, but much of it was plagiarism. Using this information, I reached out and encouraged legitimate use, even participating in sites and discussions that properly used my content, and decided to tackle the plagiarists.

The system worked very well for me. It let me meet many people and reach out to a whole new group of people. It also let me convert some cases of mistaken identity into legitimate uses and stop plagiarists by the hundred.

Though I eventually abandoned my site, there is no doubt that tracking my content helped me expand my audience and my understanding of it. These two things are at least as valuable, if not more so, than the copyright uses for many bloggers and smaller publishers.

However, Matthew Whittaker, the creator of MyFreeCopyright, previously covered here, has launched a new service entitled CopyrightSpot. The service aims to compete with established competitors by helping users search for their works easily and efficiently.

But how is the service going to fare in a crowded market? With established competitors and big names, how is CopyrightSpot stacking up?

I put the service to the test to see how well it performs against Google and some of its competitors. The results were surprising.

How CopyrightSpot Works

Anyone who has used Copyscape or Bitscan (prior to its conversion to Copy Alerts) will be very familiar with how to use CopyrightSpot. The user simply inputs the URL of the site or page that they want to check and then hits search, CopyrightSpot does the rest.

Currently, the service is in an alpha release state and the basic search is the only service provided. Right now there is only the search box and the results page. Both the link to the blog and to the dashboard do not work.

However, the Dashboard link provides a hint at a future “premium” service that may provide some case management systems.

It also provides a service that enables you to contact CopyrightSpot and suggest your ideas for the service.

However, the real question of any such service is not what services it provides, but how well it detect copies of original work.

How it Performs

Since CopyrightSpot is in an alpha release, I feel the need to go very easy on it. However, since many have already started relying upon it as their primary copy detection tool, I did want to put it through a few tests to see how it stacked up against both Copyscape and Google itself.

In that regard, I chose to perform a trio of tests similar to the ones I did against Copyscape last year.

First, I ran one of my poems, being careful to use a printable page to avoid false positives due to comments, through CopyrightSpot, Copyscape and Google (using a statistically improbable phrase) and compared the number of results detected for each.

Here are the results:

1. Google: 53 Results
2. Copyscape: 40 Results
3. CopyrightSpot: 5 Results

Clearly, CopyrightSpot lagged far behind the other two services. Where Google found 53 results and Copyscape 40, meaning that with likely mismatches in Google it was nearly 100% perfect, Copyright Spot only located 5 pages, perhaps as little as 10% of the potential matches.

I ran a similar test, this time using a prose work, once again using a printabe page. The results from that test were as follows:

1. Google: 19 Results
2. Copyscape: 8 Results
3. CopyrightSpot: 5 Results

This time, CopyrightSpot was much closer to Copyscape but both lagged significantly behind Google itself.

Finally, I tested the home page of my old site, an intro that I knew was regularly copied. I ran it through all three services with the following results:

1. Google: 12 Results
2. Copyscape: 19 Results
3. CopyrightSpot: 16 Results

These results seem odd to me as both Copyscape and CopyrightSpot bested Google. I tried this several times with different phrases but always with roughly the same result, Google, for some reason, returns fewer results than either of the other services.

However, even then, CopyrightSpot lags slightly behind Copyscape, but not so much as to be a worrisome.

All in all, the first result is a cause of concern, especially for poets or those that deal with a large volume content copying. Overall though, CopyrightSpot performed reasonably well in all of the testing, though there is clearly work to be done on the service.

Missing Features

CopyrightSpot clearly has a good start, but has a lot of work to do. But in addition to improving their detection algorithm, CopyrightSpot needs to play catch up with some of the features its competitors have.

My personal list of needed features would include the following:

1. Text Detection: Right now CopyrightSpot can only detect text within a URL. The ability to paste text, similar to Copyscape premium, would be a huge step forward for those that do not put all of their work on public URLs.
2. Email/RSS Alerts: Copy Alerts provides a decent alert system that works well for bloggers especially and Google Alerts works great for static content and is adding RSS support.
3. Resolution Assistance: This is where larger players such as Attributor and iCopyright are staking their claim. They are hoping to be more than just tracking and detection tools, but actually helping the user either convert the use into a legitimate license or secure removal.
4. Pairing with MyFreeCopyright Service: Other than cross-linking between the sites, CopyrightSpot makes no use of its sister service. Considering that MyFreeCopyright can parse RSS feeds, it seems like a natural pair.
5. Alternate Media Detection: Services such as Picscout provide the ability to detect images and Tineye is developing a search engine that allows users to search for copies of their work. A more holistic solution is needed for artists that work in mixed media.

However, I think that the worst crime the service is guilty of is looking and feeling incomplete. The blog does not work, links on the home page go nowhere and the entire site feels unfinished. A few hours of polish could likely help this site a great deal.

Conclusion

CopyrightSpot, for the most part, seems to be off to a decent start. Though it lags behind its competitors in many areas, much of that can be explained by its alpha status.

However, if CopyrightSpot is going to thrive in this field, it needs to realize that it is up against larger, better-established competitors and it needs to provide a service that is both unique in its offerings and better in terms of its protection.

It took Copyscape a series of upgrades to get where it is and CopyrightSpot is starting out relatively strong for most searches. However, it is clear that they have a long way to go.

Right now, I would definitely encourage users to give it a try and see if it works well for them but to not rely on it as their primary or even major copy detection service. It simply is not ready for that.

Right now is a great time to adopt a “wait and see” view on this service and observe what becomes of it. If it strengthens its offerings and improves its performance, it could be a major contender in this area, especially in the niche of “quickie” plagiarism checks.

If not, then it could end up like countless other “also ran” services that never went anywhere.

That company, a Japanese organization called Gazopa produced a technology that seemed to be very similar to Tineye at first glance.

However, after using and experimenting with the service, I can safely say that it was not built for this kind of searching and, at this time, it not ready to be used as a means to detect image copyright infringement.

How Gazopa Works

In terms of how one uses Gazopa, it actually works very similar to Tineye. The user either uploads or points the search engine to an image on the Web and Gazopa then searches the Web for similar images.

However, Gazopa does offer two additional options. First, it allows users to search for images based upon a keyword, thus making it more like a traditional image search engine, or via a drawing that the user makes on an electronic whiteboard.

If you upload an image or use a drawing, Gazopa then searches its database of approximately 50 million images (Note: Tineye currently has approximately 900 million) for images that are similar to it. But rather than using fingerprinting to perform the matching, it is based on more vague elements such as shape, color and face.

The concept is interesting and it would seem to have a lot of potential for detecting heavily modified image copying, such as images that were severely cropped or distorted.

However, after experimenting some with the service, it appears that the service is not quite ready for such use, at least not at this time.

Limitations of Gazopa

The problem with Gazopa is that, no matter what I search for, the site does not seem to return exact duplicates of the image or even modifications of it. Rather, it returns images that feel like the original, but actually have little in common.

A good example is what happens when I search for the Layered Tech logo using Gazopa. As you can see in the image below, none of the top results are copies of the logo, though many exist on the Web.

Though Gazopa returned 1000 matches, none of them, at least on the first five pages, were direct copies of the Layered Tech Logo.

However, when I punched the same image into Tineye, I was given much different results. Even though Tineye was only able to pull one result down, it was a perfect match for the original logo.

The result is that, where Gazopa produces far more results per item, most of them are questionable in terms of being a direct match. However, when you consider that Gazopa has a feature to prevent direct matches from appearing in the results, it is clear that isn’t the goal of the service.

Better Uses

At this time, Gazopa does not appear to be well-geared toward detecting image copying. Though I can see a lot of usefulness for the search engine, this doesn’t appear to be one of the better applications.

I would be more fond of this search if I were a designer looking for ideas or wanting to see if there was anything extremely similar, not identical, to a work I had created.

Likewise, with the drawing feature, I might be able to see if an idea I came up with has been done before or how unique a new logo I got really was.

It is a great idea for an image search engine, it just doesn’t fill this one function. Though that may change as the database grows, it is still barely 1/20th the site of Tineye’s limited database, it still seems as if the technology is not well geared toward this kind of use.

There is a lot to be excited about in Gazopa, but unfortunately, this kind of detection is just not what they are good at. There are many other reasons to try it out, but copy detection isn’t one of them.

Conclusions

Overall, I like Gazopa. It has a slick interface, a fast search and seems to return some interesting results. However, it doesn’t meet the needs for myself and others like me who are trying to track how their work is being used on the Web.

Still, I definitely recommend giving Gazopa a try. Not only might it work better for you, but it is just something that is fun to use and to play around with as well as something that can be useful.

It may not be the end all solution to finding copied images on the Web, but it certainly can help in other ways.

This is difficult because search engines, including image search engines, are designed to look for text, not pixels. Though you can look up the title of an image, the filename or even metadata within the image, if that information was changed by a site reusing your work, it has traditionally escaped detection.

Though the technology has existed in various forms, there has never been a search engine available to the public that could take an image and look for other ones like it. That is, until Tineye.

Tineye works differently than any other image search engine. It doesn’t ask you for words or even a description. Instead, you upload an image and it returns results similar to that picture. It is fast, easy to use and, most importantly, effective.

However, there are limitations to Tineye, especially in its current form. Though artists have many reasons to celebrate, the dancing likely won’t commence for some time.

How Tineye Works

For the purpose of this demonstration, I am going to use a standard Google Logo, specifically, this image:

First, after accessing your Tineye account, you upload the image from your computer to the service.

Tineye then converts the image into a fingerprint and begins matching that fingerprint against others in its database, which currently has over 700 million images.

After it is done, Tineye returns the results, starting with the images most similar to the one you submitted, for example, the image to the left. In this case, Tineye found over 3000 matching images, the first one being an exact copy of the image I had used.

However, the real magic of Tineye is not in its ability to detect images that are identical, but to detect those that are similar, but altered. This includes images that have been resized, cropped, edited or otherwise changed. As long as enough of the original work is left behind for Tineye to understand what it is, it can report the altered version.

As you can see below, in a screen capture from page 23 of the results, that often includes very heavily altered versions of the original work.

In addition to helping you find altered version of your original image, Tineye also helps you see what was changed. For each image you see, you’re able to do a comparison where you can flip back and forth between your image and the one on the Web, noting both similarities and differences easily.

Also, from the search results, you can visit the URL the image is located on, making it easy to follow through and, if appropriate, take action against any infringement.

The site also offers a Firefox/IE plugin that allows users to perform Tineye searches from any page on the Web, thus eliminating the need to download the image first.

Why this is Important

To be fair, Tineye is not the first to attempt and succeed at this kind of matching. Other companies, including both Digimarc and Picscout, have long offered similar matching services that work without text.

However, Tineye is the first to offer a robust image matching service that is free for everyone (at least as of this writing) and is simple enough to use so that artists can take advantage of it on a whim. There is no watermarking, no technology to apply to your images, just a simple upload and search.

As I see it, this has three potential implications that are both very large and very welcome:

1. Copyright Protection: The most clear use is for artists to punch their images into the service and receive results, thus enabling them to track down potential infringements of their work. They can then take action to secure removal of the images or request attribution.
2. Image Tracking: Some images, including buttons and banners, are put on the Web with the intention of them being shared and passed around. Tineye can track the effectiveness of such a campaign and determine how many sites are displaying the image in question.
3. Orphan Works Protection: Assuming that the current orphan works legislation gets passed either as is or with only a few modifications, finding a way to search for visual work is critical. Tineye can do that. If one found a work that they thought might be an orphan, they could run it through Tineye, even scanning it in if necessary, and search for copies of it on the Web, letting them track down the copyright holder. If such a tool were effective, any qualifying search would almost certainly require such an effort be made.

In short, Tineye can help bring visual artists up on par with writers in tracking their content and being able to have their work easily searched. For this reason, Tineye has already garnered several big name clients, including the Associated Press, Digg and more.

Limitations

Of course, as with any new service, there are limitations to how effective it is. However, in Tineye’s case, those limitations appear to only be temporary and should be fixed as the service grows in size and adds features.

1. Limited Index Size: Currently, the Tineye database is at about 700 million images. While that is an impressive number, one has to remember that Photobucket alone has over 5 billion images according to their numbers. The site does not seem to detect duplications on Photobucket, Flickr or other popular image sharing sites, focusing instead on blogs. Thus, many images that are known to have many copies return no results. Though Tineye has stated that they are growing their database, the number in the index has not moved in the weeks I have been using the service and no indication was given as to when they would start indexing new images.
2. No Case Tracking: Currently, with Tineye, there is no way to track cases of plagiarism or copying so that they are not acted upon a second time. Though the site does a respectable job finding duplicate images, it does little to help the artist sort through the mess. The good news is that this is a feature Tineye has expressed a willingness to implement later.
3. No Alerts System: Where writers have Google Alerts and even CopyAlerts, there is currently no system in Tineye that will alert artists to new copies of their work being posted. Once again, this is a feature Tineye has expressed an interest and willingness in adding later.

In short, Tineye is not the system artists have been waiting for today, but it definitely has the potential to be that system in the near future.

If Tineye can continue growing and improving its service, it can easily solve a problem that has had artists struggling to protect their work for well over a decade.

Conclusion

Even though Tineye is a great service with tons of potential, in its current format with the existing limitations, it is little more than a preview of what is to come.

Though you should definitely consider registering for the Tineye beta, if nothing else than to pass along your thoughts to the creators, you should realize that the searches you perform will, for the most part, be ineffective. That will hopefully change soon though.

Tineye, right now, is not intended to be the solution to the problem, but rather, a preview of the solution. So if you want to search for your images and immediately find out who has copied all of your work, Tineye, right now, is not for you.

But if you want to see what might be coming down the pipe, definitely check it out.

Related Links

Arstechnica – Another test case
The Inquisitr – An overview of Tineye
Anniebee’s Posterous – An example where Tineye worked)
Daily Tech Report – Another Tineye overview

Further Discussion

1. How will you use Tineye?
2. What features would you like to see added?
3. How do you think image rippers will respond to this kind of search?

Video