Recognising numbers about swiftocr

[NMAC427]

I'm sorry that I didn't explain it anywhere correctly. If you train the NN for the characters 0-9 then you have to change internal var recognizableCharacters = "ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789" to internal var recognizableCharacters = "0123456789" in the SwiftOCR.swift file.
You should be able to ignore the separators and add them later.

func addSeperator(seperator: String, toMACAddress macAddress: String) -> String? {
    guard macAddress.characters.count == 12 else {
        return nil
    }

    //0..<6 because every MAC Adsress has 12 characters which are grouped into *6* groups
    var seperatedMACAddressArray = [String]()

    for sectionIndex in 0..<6 {
        let substringRange = macAddress.startIndex.advancedBy(sectionIndex * 2)..<macAddress.startIndex.advancedBy(sectionIndex * 2 + 2)
        let sectionSubstring = macAddress.substringWithRange(substringRange)
        seperatedMACAddressArray.append(sectionSubstring)
    }

    let macAddress = seperatedMACAddressArray.joinWithSeparator(seperator)
    return macAddress
}

SwiftOCR().recognize(myImage) {recognizedString in
    let macAddress = addSeperator("-", toMACAddress: recognizedString)
}

Let me know if it worked. Good luck :-)

[LeoSnek]

Hmm strange, when I change the recognisable characters to ABCDEF0123456789 it works very well but with the original set it doesn't pick up any single number. Why would that be?

[NMAC427]

The output of the NN looks somewhat like this [0.01, 0.5, 0.02]. As you can see 0.5 is the largest output and it's index is 1. After SwiftOCR knows that, it will look what character is at index 1 of recognizableCharacters and appends this character to recognizedString (which in the end gets returned).

If you set recognizableCharacters to "ABC" then you would get B as your result , but if you set it to "XYZ" you will get Y.

I hope this makes sense (I'm pretty bad at explaining ^^)

[LeoSnek]

Aha, this is why the order of the recognisable characters is so critical!

I got it working perfectly with text on a Word document but we're trying to read the address from a crappy quality screen with low res text. I shall persevere! Thanks.