Re: One for the programmers or maths/logic geeks

"John Hatpin" <RemoveThisjfhopkin@[EMAIL PROTECTED]
> wrote in message 
news:jn0e24tmtbn3gs4q5vfllodlp8e5smjkqj@[EMAIL PROTECTED]
> Don, I've read this through several times, coming back to it and
> trying again, but you're losing me each time.  Questions appear
> in-line:
>
> Don K wrote:
>
>>I'm just kibitzing here but how about something like this :
>>(read while waving hands wildly)
>>
>>Start by choosing a cell near the middle.
>
> Do you have to start near the middle, or does that just make it
> quicker?

To make it quicker. Choose a cell in the middle and look around it
for the biggest contiguous square.

>
>>Map the size of the largest contiguous square in each quadrant
>>around that cell. Call these first-order squares.
>
> What size squares?  are you talking about squares that go right to the
> boundaries, or just adjacent cells, or 2x2, or ...?

I'm saying look for the biggest square that touches the sample cell.

The way I read it, X and Y each go from 0-255.
Start with the cell at (127,127) and look around it.
The 1st quadrant goes from (127,127) to (255,255), etc.
Map the biggest contiguous square that touches (127,127) in the 1st quad.

Do the same for the other 3 quadrants.
Process the results to choose the biggest square considering all
quadrants.
For example
    the 1st quad might be 6x6
    the 2nd quad might be 3x3
    the 3rd quad might be 5x5
    the 4th quad might be 7x7
The largest 1st order square would be 8x8
This bit of info can be reused for all 64cells in that square without a
new search.

The largest 1st order rectangle would be 6x7

>
>>Notice that in each quadrant, the first-order square area applies
>>to every interior cell.
>
> And this is where I really get baffled - sorry ...

(Frantically waving my hands)
Do you get it now?


>
>>Next step would be to map 2nd-order squares on adjacent boundaries,
>>then 3rd-order, etc. and look for biggest rectangles. Maybe it would
>>be efficient to do a binary-split
>
> ... which is why by now I've no idea what you have in mind.


I mean the next step is to finish mapping the biggest contiguous square
associated with each cell by choosing the next sample point midway
between the previous boundary and the edges at 0 or 255 and repeating
the process until all cells have been mapped.

If it's useful, you could sort the mapping of biggest squares to pre-point
to the best candidate landing spots.

The 3rd step is to calculate the biggest rectangle that connects to a cell
by looking at the biggest square that would touch an edge along its
1st order square. (and then the outer edges after that, etc)

>>Enough hand-waving already!
>>I'm leaving it up to you to work out the details!
>
> Ach, I'll do the details!  I just need to understand your method
> first, it sounds intriguing.

All the breakthrough ideas that come to you in a dream, seem so
brilliant at the time, but not quite so logical once you wake up.

Don