Hi everyone. I've spent the last weeks working on a program which is able to read and understand text from an image file(OCR) by using a rudimentary neural network after training with a large amount of images(I made them myself, manually). It even shows a map of all the parts of the images that have the highest synaptic weights(warmer colors). It was made purely in D using the arsd library. Below is the link to it if you wish to take a look. For now it only understands upper case letters from the English alphabet. I'll be adding more over time. Cheers.
https://github.com/MuriloMir/Optical-Character-Recognition

On 12/5/2019 5:07 PM, Murilo wrote:
> Hi everyone. I've spent the last weeks working on a program which is able to read and understand text from an image file(OCR) by using a rudimentary neural network after training with a large amount of images(I made them myself, manually). It even shows a map of all the parts of the images that have the highest synaptic weights(warmer colors). It was made purely in D using the arsd library. Below is the link to it if you wish to take a look. For now it only understands upper case letters from the English alphabet. I'll be adding more over time. Cheers.
> https://github.com/MuriloMir/Optical-Character-Recognition

Nice work!

Nice work!

You might want to take a look at the MNIST database ( https://en.wikipedia.org/wiki/MNIST_database) which is freely available and commonly used to train neural network. It will also allow you to benchmark your implementation against other algorithms.

On Fri, Dec 6, 2019 at 10:07 AM Murilo via Digitalmars-d-announce < digitalmars-d-announce@puremagic.com> wrote:

> Hi everyone. I've spent the last weeks working on a program which is able to read and understand text from an image file(OCR) by using a rudimentary neural network after training with a large amount of images(I made them myself, manually). It even shows a map of all the parts of the images that have the highest synaptic weights(warmer colors). It was made purely in D using the arsd library. Below is the link to it if you wish to take a look. For now it only understands upper case letters from the English alphabet. I'll be adding more over time. Cheers. https://github.com/MuriloMir/Optical-Character-Recognition
>

On Monday, 9 December 2019 at 01:58:10 UTC, Mathias Lang wrote:
> Nice work!
>
> You might want to take a look at the MNIST database ( https://en.wikipedia.org/wiki/MNIST_database) which is freely available and commonly used to train neural network. It will also allow you to benchmark your implementation against other algorithms.

Ahh, thank you very much, I've just saved the link to this database and I've just downloaded it.

On Friday, 6 December 2019 at 01:07:38 UTC, Murilo wrote:
> Hi everyone. I've spent the last weeks working on a program which is able to read and understand text from an image file(OCR) by using a rudimentary neural network after training with a large amount of images(I made them myself, manually). It even shows a map of all the parts of the images that have the highest synaptic weights(warmer colors). It was made purely in D using the arsd library. Below is the link to it if you wish to take a look. For now it only understands upper case letters from the English alphabet. I'll be adding more over time. Cheers.
> https://github.com/MuriloMir/Optical-Character-Recognition

Hey!

This looks like great work to learn from! Is there any chance you could add a dub file to make it easy to build?

On Friday, 6 December 2019 at 03:03:54 UTC, Walter Bright wrote:
> Nice work!

Thank you very much, I've updated it now, now it contains also all lowercase letters from 'a' to 'n', but it hasn't trained in order to learn how to read them yet.

On Monday, 9 December 2019 at 01:58:10 UTC, Mathias Lang wrote:
> Nice work!
>
> You might want to take a look at the MNIST database ( https://en.wikipedia.org/wiki/MNIST_database) which is freely available and commonly used to train neural network. It will also allow you to benchmark your implementation against other algorithms.

I've taken a look at it and unfortunately it is a file with matrices, not with images, so it does not apply to what I'm doing. But thanks anyway for trying to help me. I guess I'll have to make all of the letters myself. I've already made all uppercase and I've made the lowercase from 'a' to 'n' already.

> Hey!
>
> This looks like great work to learn from! Is there any chance you could add a dub file to make it easy to build?

I don't use DUB. But it is very easy to build, just download it and compile it. There is a file called OCR_all_files.zip which contains everything, download it, unzip it and then just compile normally. It should work. The files that you need to compile are just the files generating.d and source.d.

> Hey!
>
> This looks like great work to learn from! Is there any chance you could add a dub file to make it easy to build?

Hey everyone. I've finally finished creating all .png images for all 26 letters in both uppercase and lowercase format. I've uploaded it to the github page already. Keep in mind that the lowercase letters still need to be properly resized and none of them are trained, meaning you can't read lowercase letters yet, but you can already see them or use them for something else.

https://github.com/MuriloMir/Optical-Character-Recognition