AI | computer vision | logo detection | 7 min read

How our AI Computer Vision Enhances your Consumer Knowledge

Don't forget to share this post!

blog-post-example-media

Every day the volume of visual content on the web increases massively, across all social networks and online media. Millions of new images are shared, with a huge variety of content: celebrities, advertising for larger brands, various backgrounds, different points of view, etc.

Analyzing this type of content is an important job for marketing teams. Having a global view of what is shared on the web is essential to measure, for instance, the success of an advertising campaign or the impact of an influencer on the brand’s image.

Computer vision is the most useful tool when it comes to analyzing online images at scale. To give you a quick definition, the goal of computer vision is to identify image content with computers in the same way that humans can. In this post, we’ll discover how image analysis and content identification helps monitor brand reputation, more particularly thanks to Linkfluence’s new artificial intelligence (AI) features.

Our AI team’s goal is to provide brands with the most cutting edge computer vision algorithms, directly integrated in the product. We currently analyze many characteristics in each image we process. At Linkfluence, we perform: person detection, image demographics, emotion detection, celebrity recognition, logo recognition, object detection, scene detection, NSFW classification (nudity content detection).

Let’s dive into the new features of our product!

4 computer vision features for best-in-class image analysis

Logo recognition raised to a new level

With our whole computer vision process, we are able to smoothly absorb a huge part of the web’s content and extract various elements from posts' visual data. 

In computer vision, logo recognition consists of identifying and classifying logos from an image. It is a challenging task because we can not really give a specific definition of what a logo is: there are multiple types of logos, with different shapes, fonts and styles. It usually involves a combination of letters, texts, figures and a fixed set of colors. However, a logo can sometimes be very different from this description and some may vary in colors, textures or positions.

Among the new image analysis features and recent updates, we worked on our logo recognition   algorithm and raised it to a new level. We are now able to detect more than 3,200 different logos with our various artificial intelligence tools and can potentially scale up to support an infinite number of logos! Adding a new logo to support is easy and requires only 10 to 15 given images with the logo, no re-training involved!

We designed, built and adapted our own process so it could run on all incoming data. The results of the logo recognition are accessible through Radarly along with other image analysis and text analysis results.

Our new logo recognition process can identify a large spectrum of logos in different locations. One of the most powerful features is the new ability to detect partially cropped or distorted logos (on clothes for example). 

In AI, precision is a metric representing the proportion of accurate results among all detections. Recall is a metric representing the proportion of detected items among all present items in an image. With our new logo recognition process, depending on the nature of the logo to detect, our precision varies between 90% and 99% and our recall varies between 80%and 90%.

 

Logo detection - Linkfluence

Improved scene detection capabilities

Many other updated features come with the recent product release. The scene detection analysis is now even more relevant. Scene detection in image can be used to describe the scene context of an image, and better understand the overall context of a verbatim (which can have several downstream use cases such as consumption moment recognition). We are able to recognize over 120 scenes and situations in images. By trying to adapt the process to the characteristics of the data we usually get, the scene detection improved in accuracy and in the amount of detections. 

Scene detection - Linkfluence

NSFW classification 

The NSFW (Not Safe For Work) classification is an analysis classifying the image into two categories: neutral or nudity.  Filtering nudity is not something all platforms do and a lot of spam comes from this type of content. This tool can help measure appearances in porn content or simply be seen as a filter to eliminate nudity content from the results.

This is a huge feature to remove noise and make sure you’ll get the best set of data as the basis of your analysis.

Memes recognition

Finally, one of the recent new highlights is our memes recognition algorithm. There is no clear definition of what exactly is a meme, it can be seen as an amusing image that is spread widely online (especially through social networks). Memes are very common all over the internet, especially on social media platforms. They represent an interesting view of the social opinion on a topic, a brand or a celebrity. On our end, the goal of this AI feature is to establish whether an image is a meme. This functionality could help brands monitor their impact on social media, considering a meme trend tends to show an active discussion around an idea.

Here is an example of decision getting viral meme reactions: football player Mbappe signing back to PSG instead of Real Madrid, erasing many dreams from Real supporters:

Meme recognition - Linkfluence

 

Bring your consumer intelligence program to the next level with computer vision

Enhanced computer vision capabilities create various new opportunities for brands.

Empower your brand reputation monitoring

Images are a powerful source of information. With the growing volume of visual content over the internet, we want to upgrade image analysis and computer vision to new levels and offer a variety of results for each processed image.

By using AI, our clients can access a great overview of social media and identify in which images they appear with the logo recognition tool. Along with the new improvements and the different image classifications, measuring the impact of a marketing campaign has never been so straightforward.

Identify new ambassadors

With image analysis, you will be able to identify new key opinion leaders (KOL) who don’t specifically mention your brand while sharing on their social media channels but do promote your brand and your products via images.

Understand better moments of consumption

Image analysis is a key tool when it comes to better understanding moments of consumption. What is better than seeing people in action consuming your products? It gives brands a clear view of how people consume their products and could be a huge source of inspiration for their campaigns and creative contents.

If you are looking for a powerful tool to analyze the web, you are in the right place! We would be more than happy to show you how powerful image analysis is in action, and how much it could bring your consumer knowledge to the next level. Request a demo! 



Don't forget to share this post!