Fast PDE-Based Image Analysis in Your Pocket

Abstract

The increasing computing power of modern smartphones opens the door for interesting mobile image analysis applications. In this paper, we explore the arising possibilities but also discuss remaining challenges by implementing linear and nonlinear diffusion filters as well as basic variational optic flow approaches on a modern Android smartphone. To achieve low runtimes, we present a fast method for acquiring images from the built-in camera and focus on efficient solution strategies for the arising partial differential equations (PDEs): Linear diffusion is realised by approximating a Gaussian convolution by means of an iterated box filter. For nonlinear diffusion and optic flow estimation we use the recent fast explicit diffusion (FED) solver. Our experiments on a recent smartphone show that linear/nonlinear diffusion filters can be applied in realtime/near-realtime to images of size 176×144. Computing optic flow fields of a similar resolution requires some seconds, while achieving a reasonable quality.