What Is Image Recognition and How Does It Work?

Reading Time: 7 minutes

TLDR: Image recognition has a long history, going all the way back to 1956, however, this is most likely the age when it will be seen as coming into its own. It is an incredible piece of visual artificial intelligence tech that works by analyzing an image in comparison to a learned data set so that it can “see” and interpret what is present in the visual media. Image recognition plays a strong role throughout a huge range of industries and is often an indisposable piece of technology. 

When technology historians look back at the current age, it will likely be considered as the period when image recognition came into its own.

Most of us use image recognition daily without even realizing it. Everything from barcode scanners to facial recognition on smartphone cameras relies on image recognition.  But it goes far deeper than this, AI is transforming the technology into something so powerful we are only just beginning to comprehend how far it can take us.

This article takes a deep dive into image recognition. We take a look at its history, the technologies behind it, how it is being used and what the future holds.

A brief history of image recognition

There is no single date that signals the birth of image recognition as a technology. But, one potential start date that we could choose is a seminar that took place at Dartmouth College in 1956. This seminar brought scientists from separate fields together to discuss the potential of developing machines with the ability to think. In essence, this seminar could be considered the birth of Artificial Intelligence.

However, despite early optimism, AI proved an elusive technology that serially failed to live up to expectations. Throughout the sixties and into the seventies, scientists and researchers struggled to make any meaningful progress, and with funding and optimism drying up, AI was a concept that seemed restricted to the realm of science fiction.

This all changed as computer hardware rapidly evolved from the late eighties onwards. With costs dropping and processing power soaring, rudimentary algorithms and neural networks were developed that finally allowed AI to live up to early expectations.

In terms of Image recognition, the watershed moment came in 2012. The ImageNet Large Scale Visual Recognition Challenge (ILSVRC) was when the moment occurred. The ILSVRC is an annual competition where research teams use a given data set to test image classification algorithms.

Up until 2012, the winners of the competition usually won with an error rate that hovered around 25% – 30%. This all changed in 2012 when a team of researchers from the University of Toronto, using a deep neural network called AlexNet, achieved an error rate of 16.4%.

Error rates continued to fall in the following years, and deep neural networks established themselves as the foundation for AI and image recognition tasks.

How does Image Recognition Work

At its most basic level, Image Recognition could be described as mimicry of human vision. Our vision capabilities have evolved to quickly assimilate, contextualize, and react to what we are seeing.

This is what image processing does too – Image recognition can categorize and identify the data in images and take appropriate action based on the context of the search.

In simple terms, the process of image recognition can be broken down into 3 distinct steps.

1. It all begins with a massive dataset. This could be likened to the data available in the big wide world when viewed from the curious eyes of a baby. A baby doesn’t know a dog is a dog or a tree is a tree, but they quickly learn to identify and categorize the world around them.

The dataset provides all the information necessary for the AI behind image recognition to understand the data it “sees” in images.

2. The second stage is to feed the dataset into a neural network. This then becomes the training network. In almost all instances, the type of neural network used for image processing is a convolutional neural network.

In layman’s terms, a convolutional neural network is a network that uses a series of filters to identify the data held within an image.

The picture to be scanned is “sliced” into pixel blocks that are then compared against the appropriate filters where similarities are detected. The result is flagged as a high value in the output matrix.

This is a hugely simplified take on how a convolutional neural network functions, but it does give a flavor of how the process works.

3. The final step is what it’s all about. Now, the system can be fed images that are compared against the training dataset, and results or predictions based on what the network “sees” are output.

Technologies vary from platform to platform but normally include:  

• Logo/Mark Detection – AI can be trained to instantly recognize logos, trademarks, or other unique design elements.
• Object and Scene Detection – This allows a massive range of commonly seen objects to be tagged in visual media (video as well as still images).
• Text Detection – Text that is embedded into images can be converted into readable text. Similar to OCR, which many people are familiar with, but whereas OCR is restricted to documents, Image recognition can extract readability from any suitable image.
• Visual Search – This allows users to upload images or videos that can then be compared to the dataset to find identical or similar items. This technology is often paired with logo detection to find the exact same item.
• Hologram Authentication – Security foils and hologram stickers are commonly used for product authentication. Image recognition is used in these cases to verify the reliability of these.

Case Examples – Image recognition in everyday use

The next obvious question is just what uses can image recognition be put to. Google image searches and the ability to filter phone images based on a simple text search are everyday examples of how this technology benefits us in everyday life.

But it is business that is unlocking the true potential of image processing. There are huge incentives for businesses to tap this resource. According to Statista, Facebook and Instagram users alone add over 300,000 images to these platforms each minute. In today’s world, where data can be a business’s most valuable asset, the information in images cannot be ignored.

For many businesses, the use of image recognition is now a critical part of their operations, amongst common uses are:  

Fraud and counterfeit detection and protection

Image recognition is increasingly used by brands in the fight against counterfeiting. It can identify the illicit use of logos, trademarks, or other unique design elements.

Because Visual AI can process batches of millions of images at a time, it is a powerful new tool in the fight against copyright infringement and counterfeiting. Neural networks can quickly be trained to learn any design element.

This ability to quickly learn is a relatively new development. Previously this used to be a cumbersome process that required numerous sample images, but now some visual AI systems only require a single example.

Phishing Protection

Phishing is a growing problem that costs businesses billions of pounds per year. Traditional systems rely heavily on blacklists. However, there is a fundamental problem with blacklists that leaves the whole procedure vulnerable to opportunistic “bad actors”.

The problem is simply that a blacklist needs to be current, a few hours out of date, and the door is open for new threats to slip through. Current cybersecurity techniques also rely on identifying programmed threats embedded in emails. However, in the ongoing arms race between bad actors and security professionals, the latter are largely playing a game of catch-up.

Image Recognition, when used as part of a security stack, relies on none of these methods. Rather it scans emails and web pages, looking for visual clues from high-risk elements.

Protect against pirated content

The internet is awash with illicitly streamed content. The scale of the problem has, until now, made the job of policing this a thankless and ultimately pointless task. The sheer scale of the problem was too large for existing detection technologies to cope with.

The problem has always been keeping up with the pirates, take one stream down, and in the blink of an eye, it is replaced by another or several others. Image detection can detect illegally streamed content in real-time and, for the first time, can react to pirated content faster than the pirates can react.

Social Media

Social media has rapidly grown to become an integral part of any business’s brand. However, it does have unique challenges. It also can be damaging if not policed properly. Many of these problems can be directly addressed using image recognition.

The control over what content appears on social media channels is somewhere that businesses are exposed to potentially brand-damaging and, in some cases, illegal content. Image detection technology can act as a “moderator” to ensure that no improper or unsuitable content appears on your channels.

It can also be used to assess an organization’s “social media” saturation. The ability to quickly scan and identify the content of millions of images enables businesses to monitor their social media presence.

The future of image recognition

These are only a few of the real-world uses of image recognition. There are already a multitude of other uses. But the really exciting part is just where the technology goes in the future.

Here are just a few examples of where image recognition is likely to change the way we work and play.

• Driverless cars – This seems to be a sector that has seen a lot of empty promises come and go. But image recognition is going to play an increasingly large part in ensuring this oft-touted technology becomes the norm.
• Medical and surgical – Spurred on by the recent pandemic, the rise in “remote” medical consultations has been phenomenal. Image recognition will be crucial to help accurately perform remote diagnosis.
• Smart Glasses – Google glasses were once considered the next big thing but were perhaps too ambitious for the existing technologies. With integrated image recognition, such devices will be able to fulfill the early promises. Imagine smart glasses that can tell you that the tin of tuna you just popped into your shopping cart is available for a dollar cheaper just across the road. 
• Augmented Reality – Driven on by the gaming industry, augmented reality is another sector that can hugely benefit from image recognition. Already we are starting to see games like Pokémon Go utilize the technology, but rather than being restricted to games, the future is going to see this become an important tool in fields like medicine, fashion, and education.

Summing up

The importance of image recognition technology cannot be overstated. Despite being a relatively new technology, it is already in widespread use for both business and personal purposes.

Companies like Google, Amazon, Microsoft, and we here at VISUA are pushing the boundaries of this technology, and the exponential rise of computer vision is only just beginning.

There are plenty more articles that take an in-depth look at the subject on our website, including this excellent article that goes into the AI powering the Visual-AI platform in greater detail.

Book A Demo