algorithm - Robust Line Extraction from Image -

I need to remove all wall edges (including floor, wall intersection and wall, door intersections) from the image below. If I use the discovery (probability) to detect and change the information. It gives me many unnecessary and unnecessary lines I was seeing that I can refine the hard work before moving on the honey transformation.

Input image

By following the following cannie detection algorithms The given image is - I use the mean parameters as 020 minutes and maximum thresholds. I can not use too much value for maximum threshold otherwise I will lose the wall edges, but the shield will be less than the rest of the image.

I have a high density cluster of points in the window Think about identifying and if it is above some threshold then set them to zero.

P> Here is a magnificent image to be obtained following you can see the wall edges can be protected.

Can someone recommend me a better way to deal with this problem? I mean to refine the image of the idol so that I can identify the clusters of random points and remove them, but they can set them to zero. I was thinking of checking the collinear points in a window, but I do not know how will take effect? Any comments will be welcomed

I think you can filter the longest and almost vertical line Can, See this.

SimpleCV is just a shortcut library with OpenCV functions, you do not have to use it. I do not think you will have problems applying the algorithm after receiving the idea.

EDIT: OK, I thought more about your problem. Setting the cluster as zero as a preprocessing step is not really bad, what about step window enhancement? I mean after receiving the second image, apply a 2 * window size, another cluster filter with the same threshold. I think you can go on it, because wall edges are difficult to cancel.

Another way, use a rectangular window (width = 5 * height) for cluster filtering, as you have vertical edges.

There is another way to play with erosion and dilation and filter out large area blobs.

Another way, look at the top of the image, only wall edges and chandeliers you can search for a white pattern horizontally, then follow its neighbors to specify the length of the connected neighbors . Then filter more ones.