Am I stupid ?

So this seems like such a simple problem.

You have a block pixels.

If the pixels around a pixel are within tolerance (t), then they are included as part of shape. if a connected pixel to the original pixel has other pixels within tolerance (t) from it, they are also included in the shape.

the block of pixels has been reduced to a simple on/off state, no color considerations necessary.

Creating a bounding box around the beginning and end of data can lead to strange exceptions, so the method suggested seems the best way.

But its sooooooooooooo slow when you get large noisy images.

I've tried doing this a few different ways.

The last I tried is dividing everything line segments, classifying them by orientation (diagonal, horiz, vert, point) and then dividing the canvas into panels of so many pixels to prevent #oflines^2 comparisons, placing them lines in and then testing for intersection of the lines in one panel at a time.

Is there another way ?