2 key problems of image recognition
Suppose we have an arbitrary filled figure of a definite size placed at the centre of the image. There are various methods (depending on details) to recognize what it is. This task is solvable in general. Now add 2 variations: different location or changing size (proportionally). Does anybody know, are these tasks solved now or still not?
dnn ssd dnn ssd dnn ssd dnn ssd dnn ssd dnn ssd
OK, SSD is based on features. So it will work if we can define this object in terms of size- and position-independent features. Also need an efficient algorithm of their detection for many possible objects.
Now what about DNN? How can they do it?
just try it ? here are the data urls
yolo3 paperopencv wiki
well, that was "object detection & recognition". for simple shapes, there's a module for that also there are fourier descriptors in contrib, also the "1$ shape recognition" can be useful