My algorithm that is based on "philosophy of breaking complex things into smaller pieces" to compare them at their base level.
It also utilizes levenshtein distance to compare two data set similarity.
This project takes in a large data set of 28x28 images, then emits compiled comparable model files.
Mnist dataset is taken as a prime example in this repository, that contains a large 28x28 data of 0-9 numbers.
After compiling the Mnist dataset into models, we use the compare feature to effectively mimic a small OCR. (Thus its also some kind of ML)
Then it does pixel simplification and blob mapping (blue part is blob detection overlay):
Out of 6 characters in the image, it got 5 of them right.