What Algorithms are used?

[getSwiftly]

Hi, I am using swift OCR as a part of an academic work, can I know exactly the name of the NN and any other methods or algorithms used and the swift file in which they are? So that I can study them and research on their working. Thanks in advance.

[NMAC427]

SwiftOCR is relatively basic. It performs these steps:

• The input image gets preprocessed (SwiftOCR. preprocessImageForOCR()). I tried out many different approaches preprocessing algorithms and implemented the one that worked best for my use case.
• The function SwiftOCR. extractBlobs() tries to extract the individual characters (aka blobs) from the preprocessed image. To do this it performs Connected-component labelling (SwiftOCR.swift Line 238 - 328). I then filter and merge these components based on different rules. This is done to reduce noise and to combine the individual components into the bounding boxes of individual letters. I then take these bounding boxes and crop the image (SwiftOCR.swift Line 456) using them. The function then returns an array containing the images and their corresponding bounding boxes.
• I then feed these images into a NN (perceptron, 1 hidden layer, 321 inputs, 100 hidden nodes, 36 output nodes) to classify which character is visible in the image (SwiftOCR.swift Line 115).

I hope this helps you and if you have any questions, feel free to ask.