Cropping out an image from an existing image

Question

I would like to crop out an image from an existing image. I've taken an image and applied monochrome on it with threshold 98% using imagemagick (is this doable in openCV?)

The resulting image is this:

Now from this Image I would like to crop out another image so that the final image looks like this:

Question How can I do this in OpenCV? Note, the only reason I want to crop the image is so that I can use this answer to get the part of the text. If there is no need to crop out a new image and instead just concentrate on black part of the image to begin with, that would be great.

Answer 1

If the text at the top and at the bottom are the regions that you want to crop out, if they are always at the same location the solution is easy: just set a ROI that ignores those areas:

#include <cv.h>
#include <highgui.h>

int main(int argc, char* argv[])
{
    cv::Mat img = cv::imread(argv[1]);
    if (img.empty())
    {
        std::cout << "!!! imread() failed to open target image" << std::endl;
        return -1;        
    }

    /* Set Region of Interest */

    int offset_x = 129;
    int offset_y = 129;

    cv::Rect roi;
    roi.x = offset_x;
    roi.y = offset_y;
    roi.width = img.size().width - (offset_x*2);
    roi.height = img.size().height - (offset_y*2);

    /* Crop the original image to the defined ROI */

    cv::Mat crop = img(roi);
    cv::imshow("crop", crop);
    cv::waitKey(0);

    cv::imwrite("noises_cropped.png", crop);

    return 0;
}

Output image:

If the position of the black rectangle, which is your area of interest, is not present on a fixed location then you might want to check out another approach: use the rectangle detection technique:

On the output above, the area you are interested will be 2nd largest rectangle in the image.

On a side note, if you plan to isolate the text later, a simple cv::erode() could remove all the noises in that image so you are left with the white box & text. Another technique to remove noises is to use cv::medianBlur().You can also explore cv::morphologyEx() to do that trick:

cv::Mat kernel = cv::getStructuringElement(cv::MORPH_ELLIPSE, cv::Size(7, 7), cv::Point(3, 3));
cv::morphologyEx(src, src, cv::MORPH_ELLIPSE, kernel);    

A proper solution might even be a combination of these 3. I've demonstrated a little bit of that on Extract hand bones from X-ray image.

Answer 2

A simple solution: scan lines from top down, bottom up, left-right and right-left. Terminate when the number of dark pixels in the line exceeds 50% of the total amount of pixels in the line. This will give you the xmin, xmax, ymin, ymax coordinates to bound your cropping rectangle.