Image recognition improves as more use cases develop

Content reigns supreme in the media industry. Companies are always on the lookout for ways to improve their outreach and, over the last few years, artificial intelligence has played a significant role in engaging audiences.

One of the areas it has been most effective is through visuals. More than 80 per cent of the internet is image and video led, and AI is helping brands cut through the noise and find imagery at a faster rate than ever before.

Image recognition comes in many forms, from identifying someone’s face in a photo or video to self-driving cars making sense of the obstacles they encounter on the road. Thanks to machine learning and neural networks, AI recognizes images with more depth and clarity.

The machines are learning

Machine learning is helping to rewrite the rulebook on how to understand objects or the minute intricacies of a person’s face. The tech identifies vital features like eyes, nose, or even the rough edges of an object and places them into its machine learning model. By doing so, it can better understand the makeup of an image.

In the past, tasking the tech to identify images was anything but easy. Humans are taught to interpret visuals from a young age. For machines, unfortunately it isn’t quite as straightforward, and learning images could often be an arduous task.

The introduction of convolutional neural networks (CNN) is a method which has changed the landscape for image recognition. It pays more significant attention to pixels that are placed next to each other than ones further apart.

The reason for this is because the closer images are together, the more likely they are to be related. The result sees CNN matching pictures to people or objects at a faster and more accurate rate.

Image recognition in the media and brands

Many large brands are turning to machine learning to improve their image recognition process. Sky News, the international news channel, used facial recognition for the royal wedding between Prince Harry and Meghan Markle.

Using Amazon’s Rekognition software, the channel offered an online service to viewers of the wedding so they could identify which celebrities are in attendance, with individual’s names appearing in subtitles for the live broadcast. The service is named Who’s Who Live.

Apple’s latest smartphone, iPhone X, is the first of its kind to use facial recognition as a way for people to unlock their phone. Replacing the popular Touch Id, the tech uses light projectors and sensors to identify someone’s face. It builds a detailed map of your face which, in theory, provides more security than Touch Id.

There are a staggering 1.8-billion images on social media. Trying to find people, places and objects is far from easy. As the demand increases for user-generated content (UGC), so does the need to source authentic imagery.

Image recognition helps to cut through the noise and affords brands and agencies the opportunity to find content with ease. There are UGC marketplaces that use image recognition to collate these images in one place, making it easier for those who need it to find the right content.

From the face to the streets

Self-driving cars are now a reality and, by 2020, we are likely to start seeing automated vehicles on the road. By 2040, it is estimated 95 per cent of new vehicles will be autonomous. You could argue that the stakes are considerably higher for getting it right with self-driving cars.

If autonomous vehicles are unable to identify people or objects while they are moving, the results will be catastrophic. For this reason, image recognition plays a vital role in making sure driverless cars get on the road without any issues.

For the vehicles’ image recognition to work, they are fitted with several sensors to help recognise obstacles. Many of these potential objects include other moving vehicles, bicycles and, of course, people.

Removing bias

Even if it’s merely from a subconscious point of view, all humans contain bias. At times it can be hard for us to judge a situation without having some form of emotional attachment. This is where AI excels, as it doesn’t have the same emotional connection that we form.  

Image tagging plays its own role in the process of creating more diversity across different forms of media. The Bechdel test is a user-edited database that asks whether a work of fiction features at least two women talking to each other about something that doesn’t involve a man.

Image recognition means it is now possible to apply the Bechdel test to any work ever created. This is helpful to look back at past mistakes that have been made and see where certain aspects can be rectified in the future.

This is especially prevalent in the movie industry, where recent scandals have brought many of the industries’ problems to the fore. With it now easier to expose bias, the media industry is better suited to create work that is more representative of everyone and affect change.

Judging without bias

AI is not the all-seeing, all-doing machine that some have painted it as. For the tech to function correctly, it needs to have the right people feeding correct information. Otherwise, the whole idea becomes a fluid concept that doesn’t really mean much.

There have been issues in the past where AI hasn’t hit the mark with image recognition. However, more focus on potential issues and improvement in tech means that these issues become less frequent.  

What will be left is an image recognition that allows people to find the perfect needle in the world’s largest haystack.

Like this article? Take a second to support us on Patreon!