Archive for July, 2006

Alex Dragulescu is a Romanian visual
artist whose practice embraces both traditional and new media. His
projects are experiments and explorations of algorithms, computational
models, simulations and information visualizations that involve data
derived from databases, spam emails, blogs and video game assets.

via GoogleOS

Suppose you have to do an image transform like rotation, polygon mapping or some other effect that maps a xd, yd location in the destination image to a xs, ys coordinate in the source image. Sampling one pixel (nearest neighbour) is not enough to give quality results, of course. The correct way to do it is to map a square (xd, yd to xd+1, yd+1) to a shape in the source image and then sample the overlapping pixels with weigths computed from the intersection area – pretty complicated.
A much easier way to obtain high quality results is to process using a higher resolution destination image using the nearest neighbour sampling method and then downsample the image to the desired resolution using supersampling filtering.

Recently I had to do something similar to this paper – so I started inpiring from there for detecting the lines. Unfortunately I have lost about one day and afterwards sorted out that they had a weird variation of the Hough transform.

Basically they map θ = atan2(Gy,Gx) and ρ to the Hough space. This is terrible because the extracted gradients don’t have much precision (being extracted from 3×3-kernel convolution) to accurately describe the orientation.

The real Hough browses all the angles (0-180) and maps this angle along with d (distance from origin to a line with that angle to Ox and passing through the current point).

Oh well – it was wonderful, I just love them.

Update: of course we went to the after-party: