Kurdish/Arabic OCR Training

Añadido: 2026-05-06

[00:00:02]Welcome to the training session of OCR.

[00:00:05]Uh the language is Kuresh Arabic.

[00:00:08]Basically most is the Kuresh one. Um so those who know uh the Arabic text can do this project as well. So it's not just that the natives can do. So anyone who knows the Arabic text can do this project. Okay. So you will be given the project is very straightforward, very simple. You'll be given ID and password.

[00:00:29]Once you log in, you will come to this page and then you have to click on this orange button and you will come to the uh the interface where the tasks are mentioned. You have to click on you can translate this whole page but uh because I know which uh option is where. So you click on this second option and then the first one. This is labeling task.

[00:00:55]Here you can see uh the the the tasks.

[00:00:59]Um there are around 120 packets currently available. Uh to to get the task, you have to click on this first button. This is uh to label to mark. Once you click on it, you will come to this page where uh one task have five images. Okay. As you can see that one task have uh 1 2 3 4 5 okay five images.

[00:01:28]Before that let's just go to some basic rules. Uh we have to make boxes and then we have to transcribe. Sometimes the AI transcribe it very best and sometimes you have to transcribe it. So uh how you make a box? So one box for one text one line. Okay. So for example this is one line one line but as you can see there is space between this text and this text. So we will make two different boxes. So this first point is make different box number one when space is more than two spaces or two characters.

[00:02:03]Okay. Number second, when the font is different, but just because this is all Kurdish language, so you'll rarely see a difference in the font. But just remember that if the font is different, for example, there was this one this one uh um word is written in one font and then this other word is written in another font, then you have to make two different boxes. But this is very rare in this uh in this project. But again, the rule is rules, so you have to know it. Number third is that if the row is different. So for example, you cannot make whole one box for all these eight rows. Okay. So you have to make different row for a different uh sorry different box for a different row. Next is uh yeah this should come before all the making of the box but uh you have to invalidate a image. So for example this is the invalid button. So there are two conditions upon which you can make a um uh an image a task invalid. Number one if this whole image is upside down for example if it was like this wait if in the beginning from the beginning sorry from the beginning you got the image like this. So you're not going to rotate it what you are going to do is that you just need to click on invalid and give the reason that the image is upside down. Right. So just because this image was uh in the correct uh sense in the correct uh orientation so we will transcribe. Number second is that if the target language is less than 50%. So for example in this let me just show you there there can be Arabic there can be uh um English. So for example as you can see this whole text. So this is uh Kuresh language, this is English and then again this is Kuresh language. But for example, if English was more if the English lines were three lines and the Kuresh line were only one line. Okay. So then we are not going to transcribe it.

[00:04:05]We're just going to make it invalid. I hope this this is clear. So for example, if the whole text uh is not 50% the Kuresh language the the your target language, so just mark it invalid. and otherwise you have to transcribe it. The last rule is that regarding the blur text. So blur text or uh cut off text.

[00:04:29]So there are two options. One is this is this a static and the other is these hash tags. So for example only one character is uh you cannot see it's it's blur. Okay. So let me just show you what do I mean by blur. So for example as you can see there is some text here. Okay.

[00:04:54]So with the use of a mouse scroller wait.

[00:04:58]So mouse uh is very very important. A detached mouse is very important in this whole project. So it will be very difficult for you if you don't have a detached mouse because then what then it will take a lot of your time. So for example as you can see there is some reflection here in this area. Okay there was some letter but because there was a reflection of light from the back we cannot see but we have to make a box here. We have to make a box here and in that box just because the whole box is not visible we have to write hash these hashes three hashes. But as you can see in this uh this box only this is a bit nonvisible this but still if I zoom in I can see that it is a gene okay so I can uh transcribe but if you cannot transcribe it so for example let me just click on this uh yeah so uh the obviously this image was already done so I'm not going to I'll just do it do a new image for you guys. But I just want to show you for example this G was not visible. So what are I was going to write is that I should write the brackets and the historic instead of this gene. Okay, because G was not visible but now it's obviously it's a bit visible. It's 70% visible and 30% not visible but I know what is it's it's a gene here. But at this point I don't know what is this. So then I am going to write the hashes.

[00:06:40]Okay. So the whole point was that this is used when you are replacing a character when the character is uh not visible and this is used when the whole line is not visible. Okay, I think this is pretty simple and easy guidelines.

[00:06:57]Let's just start one uh one task. So this one I'll do.

[00:07:08]Okay. For making the boxes you have 1 2 3 options. One is rectangle. The other is um quadrer and the uh the third one is polygon. Okay. So the rectangle you can use. So rectangle is used for straightforward text. For example, if a text is straightforward like this. Okay.

[00:07:28]I'll delete this one. Wait, let me just delete it.

[00:07:34]And yeah, so I'll delete because there's no Okay, this one is used for these uh the text which are in a curve, not curve, which are um not horizontal or vertical, right? So for example like this one as you can see this text should have been made like made using this quadrer. Okay.

[00:07:58]So, quadraser is made like this. You have to make sure that all the text is in the box. Not even one dot is missing.

[00:08:07]Okay. So, now I did it.

[00:08:13]Okay. Uh now, how to transcribe it? Uh first of all, you have to select which language is it? It's a Kurdish language.

[00:08:20]So, I'll click on Kuresh. Okay. Now, how to transcribe it? Uh we are not going to use like type it. we are we have to use a quicker way. So what is a quicker way?

[00:08:31]It should have been for now I will write just ABC's for for the sake of it that it gets filled. Okay. After that what I'll do is that I'll just hide it from here. I'll take a screenshot of this text.

[00:08:49]I'll I uh I'll have my Gemini open already open and I'll give a prompt which I'll just show you guys now.

[00:09:01]I'll give you screenshot of text in Kuresh. You can you have to write Arabic because uh we being non-native we cannot identify which is Kuresh and which is uh which is uh which is Arabic.

[00:09:17]So we have to you take the help of yeah Kesh Arabic.

[00:09:22]You have to transcribe it as it is.

[00:09:29]Uh also tell me if it is Kuresh or Arabic. And one more thing also give me the number of total characters in the transcribed text. This is required by the client. Okay.

[00:09:55]Okay. Now I'll give it. And now I have to paste the screenshot.

[00:10:05]And now it's ready to transcribe it for me. Also, I'll have to open up uh Kuresh keyboard.

[00:10:15]Okay, this is Kesh keyboard for the sake of just making it.

[00:10:21]Okay, just copy it once. Paste it here.

[00:10:25]I just need to make sure that okay, all the words are there and then I'll have to paste it here. Match it up it a bit.

[00:10:33]Okay, the the raw this this letter ray, it has an arrow in it. And as you can see the the letter which the AI transcribe does not have. Okay. So sometimes what the AI do is that it makes it like messes up things. So that we have to fix it through the keyboard.

[00:10:54]So the ray I have to use this one. Okay.

[00:10:56]Let me just see there any other this y.

[00:11:01]So, let me see if it has. No, it has not. In the next ones, it might uh fix it. But I think this one it has missed.

[00:11:11]And I think that's it. This last y also have something beneath it. But I think we don't have any. Yeah, like this.

[00:11:19]Okay, now it's perfect. Now we'll just copy it and I'll paste it here. Now this is done. Okay, we are done with this.

[00:11:28]Now un like uh unhide it and you are done with this. Click on submit.

[00:11:33]Okay. This has also told us that it is a Kurdish language and the total count word counts are sorry the total characters are 24. So you have to note this down because in the progress sheet what you're going to write is the character character count.

[00:11:53]So I have not made this yet. Uh 24.

[00:11:57]Okay. your name, username, and one thing more. Sorry that I'm making this sheet now. This file name.

[00:12:08]Copy here. It's a copy button. Just copy it. You're not going to do anything.

[00:12:12]That's it. Paste it. And oops, sorry.

[00:12:15]Here.

[00:12:17]File name. So, your name.

[00:12:23]This is OCR01.

[00:12:26]And that's it. Alo. So this word count is for the sake of this and sorry uh total number of boxes. So this there was only one box. So yeah that's it.

[00:12:37]So I'm going to after noting down all the information and just checking it once I'll submit it. Moving to the next one.

[00:12:51]Okay. I'll just make a box. Again I cannot make this rectangle because rectangle is for straightforce. For example, see if I make a rectangle. See this will be such a big box. Okay. So what I will do is that I'll take the quadrice. So I'll just uh make perfect boxes. Perfect fit fixed boxes. Fit and fixed boxes. Everything is inside it.

[00:13:18]Okay. Now it's fine.

[00:13:26]Okay. Uh, is it Kurish language? Yes, it is. And for the transcription again, I'll hide it. I'll take a screenshot and I'll paste it to the Gemini.

[00:13:52]Okay. Okay. Now, I think there's something seriously wrong with the Gemini because last night it was giving us perfect um responses and I don't know what has happened to it but again I'll just copy it. I'll paste it because seriously it is simple text and then there is a left here and I'll just copy it once again. I'll check there is no uh arrows on ray and lamb.

[00:14:25]So yeah that's it. Unhide it. Copy the file name.

[00:14:31]Paste it. Uh character count I'll take it from here.

[00:14:36]uh 10 character count. Okay. And this is Kurdish language for the confirmation Kurdish language. Okay. Uh yes. So 10 boxes one and the rest I'll just copy it like this. Next. Submitted.

[00:15:08]Okay. Now, as you can see, there are a lot of um um text here. So, and yes, now we can make the rectangle. Rectangle.

[00:15:17]Making a rectangle is very easy. Okay.

[00:15:20]Actually, I don't have my detach mouse right now. So that's that's I cannot zoom in. But the best is to zoom it in and then make a box.

[00:15:33]Okay. I think um yeah before that what you can do is wait language. Select cordish language. I'll just hide it. the for uh for the images like these which has a lot of text what you can do is that take a screenshot whole screenshot and give it to charge GPT sorry Gemini chargity anything but because charity has only uh a limit of five images and uh Gemini is limitless so yes transcribe it perfectly because it is giving me wrong responses I don't know what is the issue with it transcribe it as word by word as it is. Sometimes AI messes up a lot.

[00:16:20]So you have to train it again and again till it is transcribing it for us. We will quickly make boxes.

[00:16:30]Okay. The next one first we'll make boxes.

[00:16:38]All these we are um Kesh language. So yes, again I'll say having a dash mouse is very very important for this one because uh let me just see if I can uh 60%.

[00:16:58]No, let me just do it more. 70%.

[00:17:04]Okay.

[00:17:06]Okay. Polish language. The next box we are not looking to for the transcription for now. Okay. We're just making the boxes.

[00:17:19]Okay. Curish language >> again. This lamb. See the lamb has an arrow. So we have to be very very careful here.

[00:17:28]And here another one another box. Okay. You have to select the language otherwise it will not trans it will not move forward. It will get stuck there. We have to be very careful about not missing or not cutting out the yay and the all the the dots.

[00:17:50]Kuresh.

[00:17:52]Okay.

[00:17:55]Let's see. Okay. Now it has transcribed us it for us. We have to just uh check it cross check it. Just copy it.

[00:18:03]Copy. Uh I'll just click on this first one.

[00:18:11]Okay. Paste number second.

[00:18:14]Okay. Uh again so copy and pasting and then then that's we are done with it.

[00:18:19]Okay. One thing more that I want to show for example if you say that okay this box it is cutting out the noon hair. So there there is this hand tool. Click on it. Click on the box that you want to change it or like reduce it or um increase it. So click on it and then you can increase and reduce. See you can do it vertically. You don't need to make the box from the scratch again. For example, if it it gets uh messed up.

[00:18:43]Okay. So one thing was this. Another thing that I want to show you guys was that for uh yes. So here this one as you can see this is these are numbers. Okay. And this is English. So when you make a box for this you have to select the English tag for it English or other language.

[00:19:06]Here it is numbers. So you have to make for example if I let let's make it here.

[00:19:10]Oh sorry I have to select the box.

[00:19:13]Make a box.

[00:19:18]Now as you can see I have to select English language here. Okay. If there are digits you have to select digit. If there is Chinese character you have to select Chinese character. If there is Arabic language or Udu language or any other similar language, this we will get to know from the AI. The AI will help us out that if it's uh uh if it's Pashto, if it's Persian, if it's Arabic, if it's Udu because all these the the text is uh very much same all these letter are there in all these languages. So if there is any other language then you have to click on other specific symbol the symbols like uh the currency symbol or at the rate that you have to uh click on the specific symbol tag. If there's a QR code you have to transcribe that and you have to click on QR code. If there's a barcode you have to click on the barcode. If for example the point that I told you here that if uh yeah the the blur text so the blur text the the tag uh for the whole line it will be unsure and if there is a punctuation then you have to um click on the punctuation.

[00:20:35]Okay. There's one mistake that I did here was that as you can see these are two different there should be two different boxes. So I'll fix it by clicking on it and then uh dragging it because the space between them is more than two characters. So this will be one box. Okay. I'll select digit sorry English and then I'll make another box and then I'll select English. Okay.

[00:21:04]Okay. And for the numbers like these. So for example, if I make a box here, I will have to select the digit. Okay, the digits because there is digit here. I just uh double check it. Yes, this is done. And then the then the rest of it. So I'm not doing the the the whole here, but I just wanted to show you how all these tags will work. Regarding the orientation here, the orientation most of these will be in orientation zero. Okay? So you're not you should not worry about when should be orientation 1 2 3 or unsure.

[00:21:42]It is when the text is um in the opposite direction.

[00:21:48]For example, wait let me just show you.

[00:21:55]So for example, orientation zero which is the normal one. It is when um the text is like this like this horizontal or like this vertical means like this A B C. Okay.

[00:22:14]This is orientation zero. Orientation one will be when this text this ABC in this form it is usually in like uh oh wait I cannot move it.

[00:22:35]Okay let me just so for example uh the text is in this form. Okay.

[00:22:44]Like this. The ABC or the text is written like this.

[00:22:49]So now this is orientation one. Okay.

[00:22:52]And if the text is written up upside down like this A B C then that's orientation two. So I think you should not worry about that because most of these are orientation zero. Regarding the handwriting um it's we we will click on true if it's handwriting means handwritten something but all these are printed so we have to select false but this these are automatically selected by the AI so you don't need to worry about this so I think I so far I have covered all once you finish all these five uh images okay you click on finish the package and submit and then you move to the next one so you keep on doing you keep on calculating your total number of boxes, right? And the total character count here. The uh the the AI will tell you the character count.

[00:23:46]Okay? One thing more and very important.

[00:23:49]It is uh telling us that some part is Kesh, other part is Arabic. So, we'll have to see which part is Arabic. See this part is Arabic. We will uh we'll have to train the the the Gemini in a way that it will give us separate separately give me transcription along with in what language it is.

[00:24:20]Let's hope it give us the the the accurate results.

[00:24:28]We're given a table format.

[00:24:40]Okay. Now it's um giving us a bit of the perfect results. So for example this one, this is now Arabic. This is not Kurdish. This is very important guys.

[00:24:51]Okay. So the above one all these these are in um Kuresh and the the the next this paragraph all of these are Arabic.

[00:25:01]One um characterization of Arabic is the use of aliflam a lot. In Arabic we use aliflam a lot but again we we are taking the help of the AI. So this one we have to select for example let me let me click on this this line. So this is not kesh this will be other. Okay. The next one this this box it will again it will be Arabic. This will not be Kurdish language. Okay. So I have to select on the other and uh this one as well. This is again so this is this is other from here I'll just copy the the text. Okay, I will uh cross check it once. I'll see if there's something that is left out.

[00:25:57]Usually the lilam and in in Kurdish most most of the time the wow and the ray that I told you that there are small arrows on it, they are missed out.

[00:26:06]Okay. And this one.

[00:26:12]Okay. And this one. Let me just double cross. I think this is fine. But let's Yes. So I think we are done with this whole u image except for the smaller one and then I can submit it. But yes uh once we are done with it the the the same procedure copy it paste it here.

[00:26:34]Count the total number of boxes. So here are the total number of boxes. 11 boxes are here. So I'll write 11 boxes. How many are the total characters that I'll get to know from here? characters are without spaces 234.

[00:26:52]>> Okay. And that's it. The the remaining that I'll copy and paste it from here.

[00:26:57]So I think um pretty simple, pretty easy project and the whole can be done through AI. Okay. So the easy way out is just take a full screenshot, train your AI bit and then you can just get started with it. Just make boxes and paste keep pasting it. Thank you guys. If you have any question do