How to detect room borders with OpenCV - opencv

Question

I want to clear a floor plan and detect walls. I found this solution but it is quite difficult understand the code.
Specially this line (how does it remove texts and others objects inside rooms?)
DeleteSmallComponents[Binarize[img, {0, .2}]];
https://mathematica.stackexchange.com/questions/19546/image-processing-floor-plan-detecting-rooms-borders-area-and-room-names-t
img = Import["http://i.stack.imgur.com/qDhl7.jpg"]
nsc = DeleteSmallComponents[Binarize[img, {0, .2}]];
m = MorphologicalTransform[nsc, {"Min", "Max"}]
How can I do the same with OpenCV?

Answer 1

In opencv there is slightly different approach to process images. In order to do some calculation you have to think in more low-level way. By low-level I mean thinking in basic image processing operations.
For example, line you showed:
DeleteSmallComponents[Binarize[img, {0, .2}]];
Could be expressed in opencv by algorithm:
binarize image
morphological open/close or simple dilation/erosion (based on what is color of objects and background):
cv::threshold(img, img, 100, 255, CV_THRESH_BINARY);
cv::dilate(img, img, cv::Mat());
cv::dilate(img, img, cv::Mat());
Further you can implement your own distance transformation, or use for example hit-and-miss routine (which as being basic is implemented in opencv) to detect corners:
cv::Mat kernel = (cv::Mat_<int>(7, 7) <<
0, 1, 0,0,0,0,0,
-1, 1, 0,0,0,0,0,
-1, 1, 0,0,0,0,0,
-1,1,0,0,0,0,0,
-1,1,0,0,0,0,0,
-1,1,1,1,1,1,1,
-1,-1,-1,-1,-1,-1,0);
cv::Mat left_down,left_up,right_down,right_up;
cv::morphologyEx(img, left_down, cv::MORPH_HITMISS, kernel);
cv::flip(kernel, kernel, 1);
cv::morphologyEx(img, right_down, cv::MORPH_HITMISS, kernel);
cv::flip(kernel, kernel, 0);
cv::morphologyEx(img, right_up, cv::MORPH_HITMISS, kernel);
cv::flip(kernel, kernel, 1);
cv::morphologyEx(img, left_up, cv::MORPH_HITMISS, kernel);
and then you will have picture like this:
One more picture with bigger dots (after single dilation):
Finally you can process coordinates of corners found to determine rooms.
EDIT: for images with "double wall lines" like:
We have to "merge" double wall lines first, so code will be visible like this:
cv::threshold(img, img, 220, 255, CV_THRESH_BINARY);
cv::dilate(img, img, cv::Mat()); //small object textures
cv::erode(img, img, cv::getStructuringElement(CV_SHAPE_RECT, cv::Size(5, 5)),cv::Point(-1,-1),2);
cv::dilate(img, img, cv::getStructuringElement(CV_SHAPE_RECT, cv::Size(5, 5)), cv::Point(-1, -1), 3);
And result image:
Sadly, if image properties change you will have to slightly change algorithm parameters. There is posibility to provide general solution, but you have to determine most of possible variants of problem and it will be more complex.