Extract grid data from noisy image in Halcon

Question

I am trying to get clean grid data from this image (don't mind the tilt, it is incidental).

This means no noisy dots and a uniform grid representing the situation. I am doing fairly well using morphology, doing:

closing_rectangle1 ()
opening_rectangle1 ()
select_shape ('rectangularity')

in different orders and amounts to clean up some noise.

It is not as complete and uniform as I'd like to achieve, so I'm wondering if Halcon has a better way of doing this (other than morphology or as an addition to) that I am not aware of. The rectangle-orientation and pattern in the grid can be different but the size is always the same.

Answer 1

read_image (Image, 'D:/OneDrive - Subpixel d.o.o/Other/Stackoverflow/5/uZKGw.png')
threshold (Image, Region, 128, 255)
connection (Region, ConnectedRegions)
select_shape_std (ConnectedRegions, SelectedRegions, 'max_area', 70)

** One region is disconnected, which interferes with watershed algorithm
dilation_circle (SelectedRegions, RegionDilation, 1.5)

get_image_size (Image, Width, Height)
gen_image_const (Image1, 'byte', Width, Height)
paint_region (RegionDilation, Image1, ImageResult, 255, 'fill')

watersheds (ImageResult, Basins, Watersheds)
select_shape (Basins, SelectedRegions1, 'area', 'and', 4000, 8000)

** Transform everything to rectangles 2 (rect1 if you can ignore the angle)
shape_trans (SelectedRegions1, RegionTrans, 'rectangle2')

** Reduce the regions a bit to achive separation
erosion_circle (RegionTrans, RegionErosion, 3.5)