By: Marta Robertson
Image recognition is going to be used on a large scale across sectors, changing the way we classify and search for visual information.
Search engines used to be all about text, keywords, and metadata. But now, we are witnessing the dawn of a new era. Image recognition is the catalyst which will change the way we see the world and look for information. Image recognition will also change the way data search works. It will make it safer, more reliable and less time-consuming. There are also potential cost savings for advertisers.
How Does Image Recognition Work?
So how does Facebook recognize you and your friends in the photo you just uploaded? It has already learned your facial features from the pictures you have tagged in the past. It is all about pattern recognition and classifications—tasks where machine learning is highly accurate.
Image recognition differs from the human brain in how the machine perceives the image. While people tend to see the big picture, computers need to break it down into meaningful fractions, analyze these and their relationships between each other.
Computer image analysis relies on feature descriptors, which isolate the main objects from the background and noise. These algorithms rely on heaps of data that serve as learning opportunities for the algorithms. The challenge is to find and tag those features during the training process, in order to teach the system recognize visual elements belonging to the same class. The sheer size of analyzed data is also a challenge. Even a medium-sized image of around 800 pixels, like your profile picture on social networks, will generate about 640K data points. As the algorithm needs about 1,000 photos to be adequately trained, that already means about 0.64 billion data points just to recognize one object, and not even with a high accuracy guaranteed.
Current Uses for Image Recognition
As this technology is the equivalent of giving eyes to the blind, its range of applications is limitless. It can be used in healthcare, education, security, automotive, aviation and more industries. Even if the algorithms are not perfect yet, there has been significant progress in detecting cancer and keeping public transport safe.
Retail & Advertising
Retail and advertising industries can benefit significantly from image recognition technology. For example, a virtual mirror can help you try on the entire collection of a brand without the hassle of bringing all the clothes to the fitting room. The same technologies can also prevent shoplifting by combining real-time data with a library of common behaviors of shoplifters.
Interactive marketing has been making an appearance since 2011, but it still amazes people and generates huge engagement, so we can expect to see more of it in connection with augmented and virtual reality.
Healthcare is a significant stakeholder in the image recognition game. Radiology is probably the first example which comes to one’s mind. X-rays are a significant source of information for doctors, and enhancing it with image recognition can be beneficial forpatients to limit the number of scans necessary.
The image recognition technology can look at CT scans and tag tumors much like Facebook tags your friends. Also, it can do this with high accuracy, speeding up diagnosis time and giving patients the opportunity to start treating their conditions much faster.
Very impressive is that image recognition can improve substantially the lives of visually impaired patients. It can “read” the contents of an online newspaper, blog or even the social media feed. With these advancements, the blind can also enjoy the little things in life.
Until now, image databases were tagged manually. Users associated keywords with images based on what they thought was important. This was a long, tedious and sometimes inaccurate process. With the help of image recognition technology, it will no longer be necessary to do this.
Stock photo and video repositories will self-tag the newly uploaded content, or at least provide authors with very relevant suggestions to choose from.
The same approach can be deployed by companies who own and process large image databases. For such organizations, having a searchable repository means more business opportunities and less frustrated customers, while also cutting workforce costs.
Such intelligent image classification methods will also help end consumers organize their photo albums on devices or in the cloud according to other criteria, not just time and location.
One of the driving forces behind image recognition is social media. By its design, it already has vast amounts of data which can be used for algorithm training. This means that it can improve and teach itself to become better at recognizing individuals, places, and objects in the photographs. This goes hand in hand with the use for the visually impaired previously described.
Future Applications of Image Recognition
If you think it’s the self-driving car, you are not wrong, but that is just scratching the surface. That is only one use, yet there are many more.
Image recognition works in perfect alignment with augmented reality, and in fact is part of the capabilities inherent to AR. Recognizing real-life objects and augmenting them with valuable information, hints and interactivity can simplify our lives while also enhancing both industrial and academic training.
Image recognition can improve safety and privacy too. This goes beyond simple biometric identification with the fingerprint. We can expect better iris recognition algorithms, as well as overall face recognition. These are on their way to becoming the standard identification methods, and we can expect them to work alongside or replace traditional character-based passwords completely.
Overall, it is safe to assume that image recognition technology will fuel a quality leap for data search and classification. This signifies a truly global trend of moving away from text and going into pattern recognition.
About the author
Marta Robertson has over 7 years of IT experience and technical proficiency as a data analyst in ETL, SQL coding, data modeling and data warehousing involved with business requirements analysis, application design, development, testing, documentation, and reporting. Implementation of the full lifecycle in data warehouses and data marts in various industries.