Object Detection for android with tesseract or OpenCV -

I have successfully integrated into my Android app and it reads whatever captures my image, but very little With accuracy. But after capturing most of the time I do not get the correct text as some lessons are also being caught around the area of ​​interest.

What I want to read is absolutely correct from the rectangular area, capturing the edges of the correct rectangle. I have done some research and have posted this on StackWorphflow twice, but right now Not even a happy result was found!

There are two terms I have made:

I'm not sure whether to go ahead with tesseract or OpenCV Use

with answers from many links and others, I think it is good to take a step back And note that there are actually two basic steps for Optical Character Recognition (OCR):

• Text Detection: This is the title and focus of your question, and this is the image To Aniykrit, which is an image.
• Text recognition: This is the place where real recognition is obtained, where localized image areas are obtained by detecting classified alphabetic and classified. This is also where the equipment like TesareAct comes into play.

  Now, there are two general settings in which the OCR applies:

  • Controlled: These are images taken from a scanner or similar nature, Where the goal is a document and things such as perspective, scale, font, orientation, background stability etc are very beautiful.
  • Uncontrolled / visible: These are more natural and wild pictures, like they were taken with the camera, where you are trying to identify street signs, shop names, etc.

    The Task is most applicable to the "controlled" setting. And in general, but especially for the scene OCR, the "re-training" TESTURE will not directly improve identities but can improve recognition.

    If you want to improve visualization, look at this; And if you are looking for improvement in visual text recognition, see this. Since you asked to find out, most stable extreme areas (MSER) ​​is used in the context of detection, which has a lot of implementation resources, e.g. See.

    Here's a text-detecting project specifically for Android:
    

    As many have noted, keep in mind that recognition is still open Research is the challenge.