Why is image recognition so critical for the future of brand protection?

20 Nov, 2018

Header Image ML blog

Advances in computing power are allowing techniques like machine learning to come to the surface. With over 40% of trademarks around the world containing an image component, it’s becoming more and more difficult to rely solely on words to describe something as visual and unique as a logo. This problem is compacted by the rate of data growth, which is expected to be 350 percent higher in 2019 compared to 2015, according to the United Nations Economic Commission for Europe. This is a growing nightmare for brands in today's environment where we’re already seeing 70% of counterfeit goods sold online and 230 million new tweets on Twitter each day. We are becoming extremely reliant on processes like machine learning in order to tackle the mammoth task of brand protection online.

This nightmare can only be solved using state-of-the-art technology, with the power to not only manage this problem, but interpret and analyse big data. To explain how, we’ll need to start with how Machine Learning is related to image recognition.

Machine Learning (ML) and Computer Vision go hand in hand. Try to think about it as a formula: Computer Vision + Machine Learning = Image Recognition.

Computer Vision on it’s own looks at pixel by pixel similarity, evaluating the noise, texture and shape of an image, whereas Machine learning allows us to train a computer to grasp the semantics of all visual elements within an image. When AI is used in computer vision, we’re able to teach a computer to see more like a humans do.

If you look at the example below, image processing techniques alone would rank the S logo high in visual similarity, however once employing machine learning, trained using a database of millions of trademarks, we would see similar looking diamonds ranked much higher. Using AI-powered image recognition will not only help grasp the semantics of this image (diamond) it will help rank similar-looking diamonds higher.

machine learning illustration 2 3

This simplified illustration, is a way to visualise an extremely complicated process or set of processes. Machine learning is made up of many factors, where many algorithms work together with many layers. Huge datasets (of good-quality data) and extreme amounts of computing power are essential. With many machine learning algorithms already publicly available, it’s important to understand the difference in the technology that we and many other AI industry leaders, such as IBM, Google and Amazon, are developing. A recent report featuring the leaders in this space, estimates the Machine Learning market will be worth 8.81 Billion USD by 2022. Our very own Machine Learning Researcher, Dr. Sivapalan who leads our computer vision team explains; “Cloudsourced API’s are useful for general requirements, however when you need to complete a specific task like searching and ranking visually similar logos, specific machine learning techniques need to be tailored and trained accordingly. My responsibility is to work with TrademarkVision’s proprietary image recognition technology and leverage our entire database of trademarks to develop machine learning algorithms to continuously improve our computer vision capabilities”.

The capabilities of AI are becoming widely known and the topic is heavily debated, however AI is already a part of many of our lives. As we watch the transformation unfold in the automotive industry, what we don’t see is the transformation that is starting to trickle down into many niche industries, such as the 1000+ searches being conducted on EU trademark search engine “eSearch” every day. Thousands of users everyday are unaware of the AI technology behind their image search.

Before we know it, AI applications will be seamlessly integrated into everything we do.

 

Back to blog