Metadata Challenges

6

June 13, 2007

Andreas Trampe, head of the photo department at DER STERN, Germany's leading general interest magazine, gave the keynote address at the "International Photo Metadata Conference of the IPTC" that was held in conjunction with the recent CEPIC Congress in Florence, Italy. He outlined the current challenges picture editors face in trying to find the right image, on deadline, and made clear why it is absolutely necessary to address, and find solutions to, the metadata problem as quickly as possible. As more and more images are being pumped into databases around the world the problem magnifies daily.



Mr. Trampe has 25 years in the business as a photographer and editor. He has worked for Stern for eleven years, and for the last eight has been one of two photo editors. His department has eleven picture researchers and eight assistants or secretaries. The magazine has offices in Paris, Rome, London, New York and Hong Kong, where another six picture researchers work.

He said, "We work every day, including weekends, and, if necessary, we'll also come in over Christmas or New Year. Unlike a few decades ago, there is simply no way we could structure our work any differently: if we leave the office at 6 p.m. on a Friday and return at 8 a.m. on the following Monday morning we would have a substantial backlog on our hands. Over the weekend our system receives around 25,000 new images. Monday morning is deadline for Stern. The editorial conference takes place at 8 a.m. and my editor-in-chief will want to know if there are any new, important photos which need to be included in that week's issue."






Published below is the full text of his powerpoint presentation. At points where he showed photos (which I do not have the capabilities to include in this story) I will provide brief descriptions in brackets [ ] to aid in understanding the text.

We import images continuously from 12 press agencies, receiving around 12,000 pictures a day. However, we only use approximately 250 photos per issue, in other words just around 0.3 per cent. Additionally, we have online access to around 300 databases and can search, download and use approximately 60 million images at any hour of the day.

Let's log on to Stern's editorial system from here and look at a few images together, than you can see, how many pictures we get on a normal workday. Here you can see we got 13449 pictures yesterday. [Many of these images are close similars.] These figures show that we no longer have the problem we had, say, up to 15 years ago of not having enough photos. The opposite is true: we now have access to such a multitude of images when researching for any given story that the search process has become time-consuming. Finding the right image, the relevant image and the most appealing image has become an extremely lengthy process.

Let's look together at few examples. Yesterday, as I am sure you all know, was the opening day of the G8 summit meeting in Heiligendamm, a small German village on the Baltic coast.

We are going to search through all the news agencies Stern uses in Hamburg; we'll enter "Heiligendamm" as our search term and won't specify a particular period of time. As you can see, this search returns 16884 results, all of which we have to view.

Not exactly effective, is it? To get better results we need to narrow down the search parameters slightly by applying some search filters, for example by defining a specific period of time. Let's search Heiligendamm only in the last 3 days.

Now we've trimmed our results down to 518 images, so that has worked quite well.

But what if you are not searching for just any old picture of Heiligendamm, but for images of the demonstrations around the G-8 Summit? So let's search for "Demonstration" as well.

Again, this has been quite effective; [2483 images] the results are concise, and the quantity is manageable, particularly if we narrow the time frame even further.

Now let's extend our task slightly. To coincide with the coverage of the G8 summit, you have been asked to do a portrait of Tony Blair. As this is the last time he'll be attending a G8 summit meeting, it is a good time to review his ten years in power. This search is a little more difficult to limit in terms of the time frame, because you need images from the past ten years of his premiership. What were the highlights of his ten years in power?

It's early in the morning, you haven't had your cup of coffee yet and the only notable events you can think of are; that he was an adamant supporter of the Northern Ireland peace process, that he sent his soldiers to fight in Iraq and that the British economy is doing quite well.

Difficult - or ! As most of the material in the Stern database is fairly recent, we will have to use online searches to find what we want. Let's start with the traditional British agency, Reuters. And because we need images from the past ten years, we'll simply enter "Tony Blair".

That was not so clever. Reuters has hit back with a flood of images, returning 30 pages, each containing 100 photos for a total of 3.000 Findresults. [Many of these images were demonstrators and signs with no Tony Blair in the picture. Some portraits were people other than Tony Blair.] This is obviously going to take some time.

Reuters and the other major image agencies, such as AFP and Gamma, have since realised that this can be a problem and have started offering packages and image libraries covering specific events. The agencies' picture editors pre-select the best images and compile them in special libraries.

Remember that so far we have only searched one database. Just imagine if we were to start searching in one of the numerous image portals (these are meta search engines covering several photo agencies). Let's start with a successful German portal, like Topixx.de. This portal brings together more than 130 photographers and small photo agencies. On the one hand, this is a good thing, because it lets us search more than 130 databases at one time. But let's look to the results: [For Tony Blair it found 1715 images.]

But, the Problem is not the sheer quantity, but the quality of the selections is often bad. A common problem in small agencies is that the photographers edit their own photos, and it can, understandably, be very difficult for them to select the best of their own pictures. The result is lots of pictures but not very many motifs.

What Improvements Are Needed

What could make life easier for the users, the agencies' clients? What could really be improved in the everyday picture business?

In technical terms, the definition, function and use of IPTC standards is common knowledge; the necessary fields have been documented umpteen times and yet, unfortunately, they are more often than not used incorrectly in everyday situations.

One of the main problems is that staff at the image agencies or photographers misuse the IPTC fields, unleashing an orgy of keywords upon us. Some slap over 100 keywords on their images to make sure that more of their work appears whenever anyone starts a search, thus increasing the likelihood that their pictures will actually be published.

Here are some funny results for Tony Blair [He showed a picture of shoes that were on a guardsman who was guarding an event which was presumably for Mr. Blair. They were not Tony Blair's shoes and who, looking for a picture of Tony Blair wants to see his shoes anyway.] The search for the keyword "Frustration" and some searches will end in total "Depression" as you may see: [The picture for 'frustration' was a shot over the shoulder of a traveller looking at an airline scheduling board. The photographer may have been frustrated, but there is no indication from looking at the picture that the traveller was. The picture for 'depression' showed two cooking pots with the wrong lids on each pot.]

You really begin to wonder: what on earth was the guy thinking of when he keyworded his pictures?!

But we have also to admit that searching for images for a specific topic can be very difficult. A depressed person will usually look quite ordinary; okay, perhaps he or she shouldn't be laughing happily on the picture. But depression, frustration, love or anger, are emotions which can be difficult to visualise. Art buyers in advertising agencies, who often need emotional images, will be more than aware of this problem.

Another very popular trick is to "rejuvenate" the images in the IPTC record 2, field 55. Simply enter today's date although the picture is an old archive image. This is usually done intentionally, as all editorial systems will display the most recent images first.

Here we got a good example, the picture shows Salvador Allende visits Fidel Castro taken in 1972 not 2003 which is the date on the picture. [Fortunately, the caption under this picture does correctly identify this date.]

But experienced photo editors are more than familiar with this trick.

And what would you think if you were asked to find photos of Florence and found these particular pictures? [The first is a shot of a statue that shows just the mail genitals. The other shows a group of colourful T-shirts.]

The truth is that many trainees are key wording pictures in the agencies, many photographers or their assistants have never received adequate training. They are largely self-taught or are trained on the job by their colleagues, who themselves have never had proper training. They search desperately for standards and guidelines - and most of them would love to be able do things better. But there's no negative example without an exception: Florence can also look like this. This is the webpage of Italian agency Grazia Neri. [This search generated 1859 images and the first page is all of people at events and soccer players.]

If you're clever and limit the search even further, then your results could look like this: [If you limit the search by looking only for Travel/Stock you do get a lot of architectural pictures and pictures of things in museums.]

Using effective search filters, to ensure that the pictures you get actually show Florence the city, and not some football player from Florence, is becoming increasingly important. And that holds particularly true for person searches.

Just over half of all Stern's image searches are for pictures of people. Let's do a search at one of the giants of the image industry, Getty Images. Our search will be for images of the German chancellor, Angela Merkel.

You can see the problem here. In addition to the large quantity of results [4,486] - which no picture researcher would ever be able to go through - the results also include a whole lot of pictures that have absolutely nothing or very little to do with Angela Merkel. [It starts out with pictures of U2 frontman Bono who may have been at a conference that Merkel attended, but she is not shown in the picture.]

This is not a Getty problem, but a general problem of online image searches. We expect the agencies to give their images a journalistic caption. The result is that when we search for images of Angela Merkel we also get a whole load of pictures that do not show her, but, for example, people demonstrating against her policy where her name is mentioned in the photo caption.

Hundreds of picture researchers are forced to go through tens of thousands of images every day, simply because the databases doesn't offer a better, more precise way of working.

Getty Images has attempted to solve this problem by letting users specify that only images that actually show Angela Merkel are to be returned. [That works much better, but still offers lots of similars of every situation and seems to indicate there has been no editing on the agency's part. This forces the buyer to spend a lot of time going through images that are of little value to anyone.]

These solutions will become more and more important in the future; in our world everything is digital and the image libraries in archives are growing by the hour.

In the future: where will I turn first to search for images? If I know in advance that the results are going to be very, very large?

How about introducing a new, specially defined and internationally communicated IPTC field called "pictured person"?

More and more freelance photographers and small agencies send their photos direct to us. In our era of flat rate broadband and ftp transfers this makes sense. But many copyright holders are worried that they will not receive their fee once the image has been published because the advertising agency, the editorial offices or the television station does not know them personally. So they write their name, their address, telephone number and bank account details in the IPTC header. Mostly in the field that the majority of offices import: the image description, IPTC record 2, field120, sometimes also in the "Special Instructions", IPTC record 2, field 40.

[This can be helpful. Sven Beckmann adds his telephone number and bank information to his images. If the editor has never heard of him he can call the photographer to get more information. But, if the field where this information is placed is one that is searched on to find images it can cause problems. One photographer Stern deals with has an account in the BMW bank of Munich. You can probably imagine what will happen the next time someone starts a search for images of BMW cars.]

The ideal solution would be if we could configure our databases so that they ignore a specific - internationally pre-defined - field during the search. This field could then be used for all this kind of information about the agencies and photographers. Perhaps one of the existing IPTC fields could be used? Or a new one entitled "Photographer's personal information"?

One thing is for certain: users, photo editors and art buyers will in future select their choice of image sources radical: who do they really want to work with?This will be necessary to ensure their survival in the day-to-day deluge of images.You are likely to work only with those agencies that fulfil your own journalistic, visual and organisational quality standards.

For the future, I would love to see the following motto becoming universally accepted: LESS IS MORE!!! Lots of attractive motifs in as few images as possible.

Otherwise our business is likely to succumb to what I will term "Googlisation". Have you ever been on page 81 of a Google search? No? Why do you think that is? Here we have the Google Findresult for "Photo Archieve".

Google did this search in 0.12 seconds. Great - or? 258,000,000 entries in Google. But I know how fast and efficient you are. Shall we say 15 seconds per website? Then this search will take you 44,791 days, or just under 122 years - but that doesn't include any coffee breaks!

Good luck!