How to recognize a photo?

Question

Algorithms

There is an image database.
You need to take a photo from the webcam and find the corresponding image in the database (folder).
How to implement this and are there any ready-made solutions?

Answer 1

Data preparation (for manipulation speed):
1. Contour clustering. To have lines and a white background; calculation of the rotation angle in the 2D plane: (for example) the maximum number of points (forming a contour) from the center of the image to any edge along a straight line should be oriented to the upper left corner;
then - "general pixel": matrix-1: 64x64.
2. Color clustering. So that there are averaged colors of the areas (grid) of the original image. First, we rotate by the angle from step 1 and then calculate the average overall color of the pixels on the sector (if we divide the lengths of the original image vertically and horizontally into 64 equal segments and connect the points with lines that form an imaginary grid, and take one such cell-rectangle) and get matrix-2: 64x64.
Comparison of matrices or hashes of matrices.
If the neural network is learning:
3. Learning the network using 2 matrices.
and search....
PS: with zero result during the search, you can use distortions of the 2D plane of the desired image. Recommended deviations: no more than 5-7% of the original.

Answer 2

learn a neural network, here is an example for finding a car www.ab-log.ru/smart-house/neural-network/photo-analysis

Answer 3

there are ready-made solutions