lstf | The aggregated TCP flows printer in Linux | TCP library

[ EDIT - Fully extended answer to provide complete solution ]

I wouldn't usually work to provide a "solution" since I feel it is well beyond a useful, reusable Q&A format... but it was an interesting problem.

The following details a base algorithm to detect potential door openings within the floor plan. It is not performance optimized or tested beyond the single case provided. Since the definition of a door has been given by OP only as "an opening of specified width", it is also susceptible to false indications. The algorithm can only detect principle, orthogonal doors.

Example Result:

The approach is as follows:

1. Invert and threshold in the input image, so that the darkest element are cast to white (full byte value).
2. Compute a contour detection, to identify the boundaries of the now white areas.
3. Filter to only select contours on an area greater that a selected threshold (thus removing text elements an noise).
4. "Walk" the selected contours to determine the nodes at which a "corner" occurs. A corner is defined as an angular change above a threshold.
5. Analyse the detected corners for pairings that qualify as "doors".
6. [Superfluous Rendering] Finally, raster within the rectangular bounds for the filtered contours, in order to white fill them into a resultant image. (Note: This isn't computational efficient or elegant, however the EmguCV methods for contour filling only support convex contours). The "doors" are also rendered in red.