This invention describes a computer method to identify objects in images, especially those that might lack many distinct visual features. It works by first finding small, recognizable elements like labels or icons within the object, using other visual points as a guide. Then, it uses the size relationships between these recognizable elements and the rest of the object to narrow down the possibilities and identify the object.
Why it matters: Filed before the widespread adoption of advanced deep learning for computer vision. Modern neural networks could automate and significantly improve the accuracy of feature point and scalar identification, making the ratio-based matching more robust and efficient than in 2016.
AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.
Reinvent this with AI