I think I've now got a method which works quite well (though I've been looking at these pictures for so long my judgement may well be awry).

The problem is that perspective correction works on a matrix which has its centre at the exact middle of the image. This is where the software imagines the vanishing point to be. Unfortunately if you shoot from an angle the vanishing point is displaced to one side ( and up/down if you are not level). It is about this point that transformation matrix should work.

So take this image (hdr lens corrected through PT lens - but otherwise as shot)

Now make a second layer and draw lines to work out the actual vanishing point

This shows the locus as being just below the lh of the window - but ill defined as a result of perspective issues.

Now copy the original image (without lines)to a new image which is more than twice the size of the original, and anchor it so that so that the centre of the new image is just below the lh of the window.

Apply perspective correction to the new image - this now has its transformation matrix centred close to the actual vanishing point so introduces less distortion.

This gives

which if you draw on the lines again shows a nicely defined vanishing point

But

- Have I fooled myself - it looks right but am I just blinded by looking too long
- Does this make sense

And why didn't I think of it ages ago?

Nick

## Comment