In 2011 Occupy Wall Street was in full swing. Neptune completed its first orbit since it was discovered. Adele released 21. Carole Evans and James O Jenkins founded Portrait Salon.

Portrait Salon would be a Salon des Refusés and celebrate images rejected from the prestigious Taylor Wessing Prize.

'Portrait Salon Website'

Last year the Taylor Wessing Prize received 4,462 images from 1,973 photographers. With each entry costing £28 it directly raised at least £124,936.

However, things have changed since 2011.

The Taylor Wessing Prize is now dwarfed by the fantastic British Journal of Photography’s Portrait of Britain competition. Founded in 2016 it received over 13,000 entries this year for one of 200 spots in a nationwide billboard exhibition.

This year Portrait Salon responded by accepting submissions that been rejected from both the Taylor Wessing Prize and Portrait of Britain to show that there is top quality photography which is not getting the exposure it deserves.

Challenges

I have been managing Portrait Salon’s website since 2013. Each year the competition changes.

Some years we have used a panel of judges, other years we have opened a limited pool of images to a public vote.

This presents a series of on-going challenges. How to accept images; how to take payment for images; how to create a platform to allow judges to select images fairly.

This year was a little different.

Every image submitted was made public and put to a public vote.

'Portrait Salon Website'

In total the competition featured 645 images submitted by 201 photographers who had been rejected by either the Taylor Wessing Prize or the Portrait of Britain.

On average people submitted 3.21 images.

Images Submitted  
1 31%
2 17%
3 10%
4 8%
5 6%
6 27%
6+ 1%

One person submitted 24 images across four entries. One person submitted the same image set on the start date and close date of the public vote.

The public vote ran for 20 days. The website consumed 70GB of bandwidth during that time, and handled over 3 million requests.

In hindsight this this maybe wasn’t the most practical idea.

If you took 1 second to review an image you would need to commit 11 minutes to fully review the whole set.

The Public Vote

In total over 15,000 votes were cast with 568 images receiving one or more votes.

Only 12% of images received no votes.

Importantly we put no explicit rules for voting for your own images in place.

We assumed that even if we stated you shouldn’t vote for your own images, some people still would. That would punish good actors who abide by the rules and rewards those that break them.

Instead, we made it difficult to find your own images. We randomized the order of the set every few hours and obfuscated the photographers name.

By making it difficult to find a specific image we forced people to review the whole set.

The Stats

Despite of this 80% of people cast one vote for a single image. Only 7.5% of people voted for more than 4 images.

Number of Different Images Voted For Per Person  
1 79%
2 9%
3 4%
Over 4 7.5%

The highest number of unique votes cast by a single person for different images was 78. This meant they voted for over 13% of the images displayed.

One for all and all for one

While we allowed people to vote for their own images we did create a mechanism to disregard aggressive vote manipulation.

We wanted to respect the time people had taken to vote and ensure the process was fair to everyone.

As people were encouraged to vote on multiple images we cannot simply discount multiple votes made by the same person (as they might be for different images).

As a rule of thumb I try to collect only the bare minimum amount of personal information necessary for a system to work.

Due to this we avoided more complex verification methods such as account creation and/or email address verification. This reduces the data we hold and lowers the barrier to vote.

Instead (without giving away the recipe for our secret sauce) we built a vote process around the principle that unique votes are more valuable than people voting multiple times for the same image.

In layman terms, if you vote for your own image 10 times, we would trust those votes less than if you had just voted for that image once. If you voted for your image 100 times we would trust those votes significantly less than if you voted for that image 10 times.

In total 56% of images had a 100% trust rating in this system.

This means more than half of images which received votes did not have multiple votes from the same person associated with them.

19% of images had a moderate trust rating, meaning they had more unique votes than manipulated votes associated with them.

25% of images had a low trust rating that implied heavy vote manipulation.

Out of those that voted multiple times for the same same image 55% voted for the same image between 2 and 5 times.

23% voted on an image between 6 and 10 times, and 22.5% voted over 10 times.

Multiple Votes Cast  
2 17%
3 18%
4 12%
5 8%
6 to 10 23%
10 to 20 12%
20+ 10.5%

Only 2% of those that voted more than once for the same image voted more than 100 times.

… and one person took the time to vote for a single image 1,576 times.

Explore past winners at Portrait Salon. This years winners will be announced in November.