yepp, thank u v much. it's a simple algorithm. after defining your bg-colour (white or transparent) an observer is following the bounding-rect ( top_left to bottom_left, ...), checks for the 1st pix != bg-col (row/column) and stores the pos in an area. i repeat that cycle 4 times, varying the starting corner. after that i get the intersection of all areas

. (i also used a contrast filter) works fine for simple surfaces + it is quite fast because it is linear
linked to my sprite-class i get a dynamic collision model (each frame gets it's mask).