Often misused, the concept of “image identification” is more and more common. By definition, image identification allows to identify and discern the nature of an image or its components. However, it does not specifically correspond to a scientific notion and it lacks precision.
Image identification does not really exist
While the use of “image identification” might not be appropriate, it does refer to computer vision. Indeed, it is thanks to computer vision that it is possible to recognize the nature of an image and what it represents, thus suggesting similar content.
More precisely, there are two distinct subcategories of computer vision allowing to perform these tasks: visual search and image recognition.
Visual search: an image is worth a thousand words
Image identification is closely related to visual search because it allows us to find the origin of an image or to propose similar or related content.
Have you ever wondered how a website could suggest you items that you could potentially like? The answer is visual search.
Indeed, e-commerce, especially in the fashion industry, is increasingly adopting visual search to provide a more personalized customer experience. It is now possible to provide customers with product suggestions by theme or style in relation to the associated product sheets previously consulted.
Visual search, or visual search engine, browses the Internet from an image. It works with all types of images such as screenshots, internets images or other pictures.
To help you understand the principle, let’s take the example of Google Images, the best known among visual search platforms.
Let’s do a simulation of how to use Google Images. First, you decide to upload the image of a red car, which you import from your computer. Next, Google will offer you a possible associated search, a set of similar images and finally pages containing images identical to yours. This list of suggestions is made according to a certain number of criteria: colors, shapes, patterns, etc.
TinEye is another example of an innovative visual search engine. Its goal is to find the origin of an image by providing you with the web pages where it appears. Its use is very simple. Simply upload your image to the search bar and then explore the list of pages where your image appears.
Image recognition: an image is worth a thousand labels
In addition to visual search, image identification often seems to be associated with image recognition. The latter, a subcategory of computer vision, consists of a set of image detection and analytics methods to automate a specific task or process.
For image recognition to be effective, it is necessary to train the neural networks of an algorithm through an annotated database. There are already trained neural networks online, which can recognize several thousand labels. For instance, these are Google Cloud – Vision API and Clarifai.
Let’s take Clarifai as an example. Through the platform, you can search by concept, image, metadata or geolocation. In this case, we will only focus on image research.
The idea is simple. The platform has an image database classified by concepts. In particular, Clarifai offers you a list of 10 different concepts to test the efficiency of the algorithm. We decided to choose the concept of “Siamese cats”. All you have to do now is upload a Siamese cat image to the search bar to get all the similar images. This result refers to the labels assigned to the different images in the database.
Conclusion
To conclude, image identification refers, in most cases, to computer vision and in particular to visual search and image recognition. However, it is important to keep in mind that it is not an established scientific concept.
Are you curious about this topic? Do you want to understand more about computer vision and image recognition?
