Just released: Improved Conceptual Image Similarity algorithm in the Global Brand Database
January 8, 2021
The newly trained algorithm identifies the figurative elements embedded in trademark logos with a higher precision. This is thanks to the smart detection of the verbal elements embedded within a logo when it is indexed.
A combined logo with both verbal and figurative elements is now better identified as very similar to the logo containing the same figurative elements but not the verbal elements. Consequently, the system sorts relevant results with a higher accuracy.
Please note that the use of the image cropping tool (within GBD) to delimit the region of interest containing the visual concept you search in your query image still remains a good practice to obtain the best results.