I'm developing an app which can recognize license plates (ANPR). The first step is to extract the licenses plates from the image. I am using OpenCV to detect the plates based on width/height ratio and this works pretty well:
But as you can see, the OCR results are pretty bad.
I am using tesseract in my Objective C (iOS) environment. These are my init variables when starting the engine:
// init the tesseract engine.
tesseract = new tesseract::TessBaseAPI();
int initRet=tesseract->Init([dataPath cStringUsingEncoding:NSUTF8StringEncoding], [language UTF8String]);
tesseract->SetVariable("tessedit_char_whitelist", "BCDFGHJKLMNPQRSTVWXYZ0123456789-");
tesseract->SetVariable("language_model_penalty_non_freq_dict_word", "1");
tesseract->SetVariable("language_model_penalty_non_dict_word ", "1");
tesseract->SetVariable("load_system_dawg", "0");
How can I improve the results? Do I need to let OpenCV do more image manipulation? Or is there something I can improve with tesseract?
Two things will fix this completely:
Remove everything which is not text from the image. You need to use some CV to find the plate area (for example by color, etc) and then mask out all of the background. You want the input to tesseract to be black and white, where text is black and everything else is white
Remove skew (as mentioned by FrankPI above). tesseract is actually supposed to work okay with skew (see "Tesseract OCR Engine" overview by R. Smith) but on the other hand it doesn't always work, especially if you have a single line as opposed to a few paragraphs. So removing skew manually first is always good, if you can do it reliably. You will probably know the exact shape of the bounding trapezoid of the plate from step 1, so this should not be too hard. In the process of removing skew, you can also remove perspective: all license plates (usually) have the same font, and if you scale them to the same (perspective-free) shape the letter shapes would be exactly the same, that would help text recognition.
Some further pointers...
Don't try to code this at first: take a really easy to OCR (ie: from directly in front, no perspective) picture of a plate, edit it in photoshop (or gimp) and run it through tesseract on the commandline. Keep editing in different ways until this works. For example: select by color (or flood select the letter shapes), fill with black, invert selection, fill with white, perspective transform so corners of plate are a rectangle, etc. Take a bunch of pictures, some harder (maybe from odd angles, etc). Do this with all of them. Once this works completely, think about how to make a CV algorithm that does the same thing you did in photoshop :)
P.S. Also, it is better to start with higher resolution image if possible. It looks like the text in your example is around 14 pixels tall. tesseract works pretty well with 12 point text at 300 dpi, this is about 50 pixels tall, and it works much better at 600 dpi. Try to make your letter size be at least 50 preferably 100 pixels.
P.P.S. Are you doing anything to train tesseract? I think you have to do that, the font here is different enough to be a problem. You probably also need something to recognize (and not penalize) dashes which will be very common in your texts, looks like in the second example "T-" is recognized as H.
I don't know tesseract too much, but I have some information about OCR. Here we go.
In an OCR task you need to be sure that, your train data has the same font that you are trying to recognize. Or if you are trying to recognize multiple fonts, be sure that you have those fonts in your train data to get best performance.
As far as I know, tesseract applies OCR in few different ways: One, you give an image which has multiple letters in it and let tesseract do the segmentation. And other, you give segmented letters to tesseract and only expect it to recognize the letter. Maybe you can try to change the one which you are using.
If you are training recognizer by yourself be sure that you have enough and equally amount of each letter in your train data.
Hope this helps.
I've been working on an iOS app, if you need to improve the results you should train tesseract OCR, this improved 90% for me. Before tranning, OCR results were pretty bad.
So, I used this gist in the past to train tesseract ORC with a licence plate font.
If you are interested, I open-sourced this project some weeks ago on github
Here is my real world example with trying out OCR from my old power meter. I would like to use your OpenCV code so that OpenCV does automatic cropping of image, and I'll do image cleaning scripts.
First image is original image (croped power meter numbers)
Second image is slightly cleaned up image in GIMP, around 50% OCR accuracy in tesseract
Third image is completely cleaned image - 100% OCR recognized without any training!
Now License Plate can be easily recognized by mlmodel. I have created the core model you can find it here . You just need to split characters in 28*28 resolution through vision framework and send this image to VNImageRequestHandler like given below-
let handler = VNImageRequestHandler(cgImage: imageUI.cgImage!, options: [:])
you will get desired results by using my core mlmodel. Use this link for better clarification but use my model for better results in license plate recognition. I have also created the mlmodel for License Plate Recognition.
Related
my first hurdle so far is that running tesseract vanilla on images of MTG cards doesn't recognize the card title (honestly that's all I need because I can use that text to pull the rest of the card info from a database) I think the issue might be having to need to train tesseract to recognize the font use in mtg cards but im wondering if it might be an issue with tesseract not looking or not detecting text in a section of the image (specifically the title.)
Edit: including an image of a MTG card for reference.http://gatherer.wizards.com/Handlers/Image.ashx?multiverseid=175263&type=card
Ok so, after asking on reddit programming forums I think I found an answer that I am going to pursue:
The training feature of tesseract is indeed for improving rates for unusual fonts, but that's probably not the reason you have low success.
The environment the text is in is not well controlled - the card background can be a texture in one of five colours plus artifacts and lands. Tesseract greyscales the image before processing, so the contrast between the text and background is not sufficient.
You could put your cards through a preprocessor which mutes coloured areas to white and enhances monotones. That should increase the contrast so tesseract can make out the characters.
If anyone still following thsi believes that above path to be the wrong one to start down, please say so.
TLDR
I believe you're on the right track, doing preprocessing.
But you will need doing both preprocessing AND training Tesseract.
Preprocessing
Basically, you want to get the title text, and only the title text, for Tesseract to read. I suggest you follow the steps below:
Identify the borders of the card.
Cut out the title area for further processing.
Convert the image to black'n'white.
Use contours to identify the exact text area, and crop it out.
Send the image you got to Tesseract.
How to create a basic preprocessing is shown in the YouTube video Automatic MTG card sorting: Part 2 - Automatic perspective correction with OpenCV Also have a look at the third part in that serie.
With this said, there is a number of problems you will encounter. How to handle split cards? Will your algorithm manage white borders? What if the card is rotated or upside-down? just to name a few.
Need for Training
But even if you manage to create a perfect preprocessing algorithm you will still have to train Tesseract. This is due to the special text font used on the cards (which happens to be different fonts depending on the age of the card!).
Consider the card "Kinjalli's Caller".
http://gatherer.wizards.com/Handlers/Image.ashx?multiverseid=435169&type=card
Note how similar the "j" is to the "i". An untrained Tesseract tend to mix them up.
Conclusion
All this considered, my answer to you is that you need to do both preprocessing of the card image AND to train Tesseract.
If you're still interested I would like to suggest that you have a look at this MTG-card reading project on GitHub. That way you don't have to reinvent the wheel.
https://github.com/klanderfri/ReadMagicCard
I have an image for mobile phone credit recharge card and I want to extract the recharge number only (the gray area) as a sequence of number that can be used to recharge the phone directly
This is a sample photo only and cannot be considered as standard, thus the rectangle area may differ in position , in the background and the card also may differ in size .The scratch area may not be fully scratched , the camera's depth and position may differ too . I read a lots and lots of papers on the internet but i can't find any thing that could be interesting and most of papers discuss detection of handwritten numbers .
Any links or algorithms names could be very useful .
You can search the papers on vehicle plate number detection with machine learning methods. Basically you need to extract the number first, you may use sobel filter to extract the vertical edges , then threshold (binary image) and morphologic operations (remove blank spaces between each vertical edge line, and connect all regions that have a high number of edges). Finally retrieve the contour and fill in the connected components with mask.
After you extract the numbers , you can use machine learning method such as neural network and svm to recognize them.
Hope it helps.
Extract the GRAY part from image and then Use Tesseract(OCR) to extract the text written on the gray image.
I think you may not find the algorithm to read from the image on the internet. Nobody will disclose that. I think, if you are a hardcore programmer you can crack that using your own code. I tried from the screenshots where the fonts were clearer and the algorithm was simple. For this, the algorithm should be complex since you are reading from photo source instead of a screenshot.
Follow the following steps:
Load the image.
Select the digits ( By contour finding and applying constraints on area and height of letters to avoid false detections). This will split the image and thus modularise the OCR operation you want to perform.
A simple K - nearest neighbour algorithm for performing the identification and classification.
If the end goal was just to make a bot, you could probably pull the text directly from the app rather than worrying about OCR, but if you want to learn more about machine learning and you haven't done them already the MNIST and CIFAR-10 datasets are fantastic places to start.
If you preprocessed your image so that yellow pixels are black and all others are white you would have a much cleaner source to work with.
If you want to push forward with Tesseract for this and the preprocessing isn't enough then you will probably have to retrain it for this font. You will need to prepare a corpus, process it similarly to how you expect your source data to look, and then use something like qt-box-editor to correct the data. This guide should be able to walk you through the basic steps of retraining.
I'm trying to create a simpler OCR enginge by using openCV. I have this image: https://dl.dropbox.com/u/63179/opencv/test-image.png
I have saved all possible characters as images and trying to detect this images in input image.
From here I need to identify the code. I have been trying matchTemplate and FAST detection. Both seem to fail (or more likely: I'm doing something wrong).
When I used the matchTemplate method I found the edges of both the input image and the reference images using Sobel. This provide a working result but the accuracy is not good enough.
When using the FAST method it seems like I cant get any interresting descriptions from the cvExtractSURF method.
Any recomendations on the best way to be able to read this kind of code?
UPDATE 1 (2012-03-20)
I have had some progress. I'm trying to find the bounding rects of the characters but the matrix font is killing me. See the samples below:
My font: https://dl.dropbox.com/u/63179/opencv/IMG_0873.PNG
My font filled in: https://dl.dropbox.com/u/63179/opencv/IMG_0875.PNG
Other font: https://dl.dropbox.com/u/63179/opencv/IMG_0874.PNG
As seen in the samples I find the bounding rects for a less complex font and if I can fill in the space between the dots in my font it also works. Is there a way to achieve this with opencv? If I can find the bounding box of each character it would be much more simple to recognize the character.
Any ideas?
Update 2 (2013-03-21)
Ok, I had some luck with finding the bounding boxes. See image:
https://dl.dropbox.com/u/63179/opencv/IMG_0891.PNG
I'm not sure where to go from here. I tried to use matchTemplate template but I guess that is not a good option in this case? I guess that is better when searching for the exact match in a bigger picture?
I tried to use surf but when I try to extract the descriptors with cvExtractSURF for each bounding box I get 0 descriptors... Any ideas?
What method would be most appropriate to use to be able to match the bounding box against a reference image?
You're going the hard way with FASt+SURF, because they were not designed for this task.
In particular, FAST detects corner-like features that are ubiquituous iin structure-from-motion but far less present in OCR.
Two suggestions:
maybe build a feature vector from the number and locations of FAST keypoints, I think that oyu can rapidly check if these features are dsicriminant enough, and if yes train a classifier from that
(the one I would choose myself) partition your image samples into smaller squares. Compute only the decsriptor of SURF for each square and concatenate all of them to form the feature vector for a given sample. Then train a classifier with these feature vectors.
Note that option 2 works with any descriptor that you can find in OpenCV (SIFT, SURF, FREAK...).
Answer to update 1
Here is a little trick that senior people taught me when I started.
On your image with the dots, you can project your binarized data to the horizontal and vertical axes.
By searching for holes (disconnections) in the projected patterns, you are likely to recover almost all the boudnig boxes in your example.
Answer to update 2
At this point, you're back the my initial answer: SURF will be of no good here.
Instead, a standard way is to binarize each bounding box (to 0 - 1 depending on background/letter), normalize the bounding boxes to a standard size, and train a classifier from here.
There are several tutorials and blog posts on the web about how to do digit recognition using neural networks or SVM's, you just have to replace digits by your letters.
Your work is almost done! Training and using a classifier is tedious but straightforward.
I get data from video so there is not way for me to rescan the image, but I can scale them if necessary.
I do have only a limited number of characters, 1234567890:, but I have no control over the dpi of the original image or the font.
I tried to train tesseract but without any visible effect, the test project is located at https://github.com/ssbarnea/tesseract-sample but the current results are really bad.
Example of original image being captured:
Example of postprocessed image for OCR:
How can I improve the OCR process in this case?
You can try to add some extra space at the edges of the image, sometimes it helps for tesseract. However, opensource OCR engines are very sensitive to the source image DPI.
I have an OCR project, but it works good only with images in which the text is fairly straight, not upside down. (not rotated text)
So I want to make OCR to be able to recognize any kind of images, even upside down. But I don't know what are approaches to solve this problem.
I need something like analysis of lines of letters, but even then I can't identify if line is upside down or not.
If the images you are performing OCR on are from a magazine or book where there is lots of text on multiple lines, I suggest trying to find the rotation of the page.
Probably the simplest way to do this is applying the hough transform for lines. Since the empty space between each line of text should be a a broad white line this could work without any preprocessing of the image. Otherwise try blurring it or using the "close" morphological operation to make the lines of text into opaque blocks.
Once you find the lines in the image with the hough transform you should just extract the principal angle of rotation (like the mean angle of all lines) and rotate it back.
My answer to you will be very high level as this is not simple, as you can imagine. You probably are doing some sort of image segmentation, where you segment each character of your text. But in order to recognize the characters, even when they are rotated, you need to use a feature vector with rotational invariant characteristics. To do it some people are using
Zernike Moment
Neocognitron neural network - widely used for handwriting
I don't think it's a simple task
Not sure if you are creating an OCR engine or using one. Most commercial OCR engines can detect that a page is upside-down (or 90 degree rotated) and auto-rotate it. For example, my company's GlyphReader OCR Engine can do that.
One simple solution is to take a portion of your image and run it through the engine at the four angles until you get back a good amount of recognized text. You can use a dictionary to see if what you are getting back is words and confidence levels to see how sure the engine is of its recognition.
If your engine can report confidence levels, and they are reporting consistently under some threshold, then you should stop and see if the document is rotated.
For 90 and 270, a hough transform will tell you whether the lines in the image are horizontal or vertical. It can also tell you if they are just slightly rotated off the horizontal so that you can correct that as well.