Tagging images is a task that, though time-consuming, is much more natural for us as human beings than it is for computers. In the last three decades a lot of research groups in universities and companies have tried to attack this problem from various angles and for different purposes. We believe we've found the right practical way to solve it for a lot of use-cases.
Machine Learning
Combining state-of-the-art machine learning approaches in Imagga's technology, allows it to be trained in the recognition of various visual objects and concepts. For each learned item the system 'sees' in an image, appropriate tags are suggested. We continuously train our technology with more and more objects and concepts, but also offer you the ability to train it with specific items, suited to your needs.
Learn more how you can build your own classifier.
Image and tags correlations
Machine learning is the first step that needs to be taken in order to suggest the most prominent and recognizable tags in an image. We've gone one step further and suggest more tags based on multiple models that we've learned. They relate the visual characteristics of the images with their associated tags in big manually created data sets.
Semantic processing
Finally, in order to further refine the quality of the suggested tags we apply multiple Natural Language Processing passes.
This technology can be also used for similarity search, via matching images that share some tags, eventually in combination with our color extraction.
Color Extraction & Search