Trying to pre process this image for OCR. And I am not able to figure out how to get rid of that bump in the centre of the captcha.
Original Captcha - Captcha
This is as far I can get but still this isn't good enough.
Processed Captcha - CaptchaProcessed
Seems to be a captcha you should solve with a neuronal network like this:
Write something to gernerate this kind of captcha by yourself (with bounding boxes) aka: training data
Train a neuronal network on your training data
profit
See how its done for a different captcha typ here: https://github.com/cracker0dks/CaptchaSolver/blob/master/docs/howToSolveNew6DigitCaptchasWalkthrough.md
Related
I'm struggling with a captcha image that has reflected letters on the bottom. Taking the shadow off the captcha, it can easily be resolved. I have tried using OpenCV, but no success so far. For example, if the captcha text is "h8mb", it will have a reflected shadow with the "h8mb" text cropped as shown below:
I will link the images below, since I can't add them due to the fact that my account is new.
https://i.stack.imgur.com/X7vG4.jpg
https://i.stack.imgur.com/YGufm.jpg
https://i.stack.imgur.com/N3GWB.jpg
https://i.stack.imgur.com/0q7hu.jpg
https://i.stack.imgur.com/yXAWw.jpg
Could someone explain to me how this could be done? Or how could I use OpenCV to do this reflection removal?
As is often the case, the design of this captcha is pretty poor.
Every sample is made of the word struck out, forming a single, largest blob. The other pieces make distinct and smaller blobs, which are too easy to erase by seed filling.
I'm trying to implement Invisible reCAPTCHA by Google. But after 5-6 times the images challenge comes up which is against the requirement . Is there a possibility of skipping the images challenge?
Not that I know of. The entire point of the invisible captcha is that it does not show up when it is sure that the page is being visited by a human. If it is not sure or certain that it is a bot, the challange is presented.
I am doing a project,where I need to determine if the image is self taken or not? I am confused whether to use machine learning approach or image processing ? Machine learning approach may not give me very effective results.I am looking for image processing method.Can anyone help me out here..?? I can detect the faces in the image but that is not enough.
Machine learning will work way better in my opinion. With few exemples you should get a correct result. But it won't be perfect.
If you want something more occurate you will need other detailed. For exemple if it is a smartphone application you know if the front camera was used. Maybe another great source of data is the accelerometer because selfies are not taken the same way of regular picture.
If you have faces in the picture and another relevant indicator you probably can state that the picture is a selfie.
if you were watching the State of the Union Address (http://www.whitehouse.gov/state-of-the-union-2013) you would have seen graphic supplements that appeared alongside of the video stream of the President that served to illustrate his key points.
The video on the site is a composite of this, but during the live streaming these were handled separately.
My question is: what is the best approach for doing this? especially if one wanted very tight control of the appearance of the graphics (i.e. right when the point is made, not before and not long after).
I'm wondering if any tools exist to facilitate this? I've been scouring google, but I don't think that I have the correct technical vocabulary for what I'm describing because I'm coming up blank.
I imagine AJAX would be a good starting point, but I'm not sure how to achieve the level of control that they had, or how to handle the back end of things.
For anyone who might encounter this challenge we devised two ways to solve it:
The first is a bit mickey mouse: It requires that you know how many images, etc you want to use beforehand (which in most cases you would). We wrote a script to repeatedly request an image and inserts it into the page, and on finding an image then request the next image in the chain.
Ie. Display default image -> request image 1
then, displaying image 1 -> request image 2
etc
From your end you can simply drop the images into a folder on your server when you are ready for them to go in. An advantage of this is that the images can be interactive, with links to other content, etc.
The big disadvantage, of course, is a lot of unnecessary requests to your page. In our case we anticipated enough traffic that it didn't seem wise. Also, there are plenty of opportunities for mistakes and depending how frequently your timer fires there are likely to be timing discrepancies.
The Second costs money: we found the program Ustream (http://www.ustream.tv/producer) which allows us all the image control we require in terms of timing with the advantage of providing support for media clips etc. And it allows you to record everything streamed.
The disadvantage is that what the user sees is an integrated video on your site, so that you have to handle links to related content and provide images (if you want your users to have access to them) separately.
Hope this comes in handy for someone
I would still welcome any suggestions on how to make the first method more effective
I am able to utilize imagemagick to properly generate animated images and I can post them to twitter without issue, however upon posting them, it seems as though Twitter is somehow destroying the animation component I'm guessing with it's reencoding of the image.
This is less than desirable in my situation as I need to post a statistical compilation of images daily to an account and need the animation to retain integrity. I am supposing that this is a function of Photobucket that they're now using or such.
How does one encode an image and upload it such that it retains its integrity? I have wondered about uploading directly to TwitPic or other options or perhaps exploring more fully the imagemagick encoding options so that they line up precisely with Twitter requirements in order to produce an image that requires no reencoding, however I'm looking for help in this regard.