An arm neon optimised image abstraction method utilising the cosine integral image method

In this paper we present a fast implementation of an automatic non-photorealistic image processing technique which transforms an input image frame of a video stream into a non-photorealistic abstracted cartoon stylised render. The approach presented utilises a fast cosine integral image method to create a separable bilateral filtering stage which operates in constant time. This is subsequently put through a colour quantisation stage and combined with an edge overlay system to generate the abstracted image output. The algorithm is implemented with OpenCV on a Beagleboard-xM running Angstrom GNU/Linux to demonstrate the improved performance obtained utilising the cosine integral image bilateral filter over the OpenCV standard bilateral filter implementation, and to demonstrate further performance improvements can be obtained through utilising optimised routines on the ARM NEON floating point unit of the Beagleboard-xM.