In this paper, we present a local adaptive filter for fast edge-preserving smoothing, a so-called cross-based filter.
The filter is mainly built on upright crosses and captures the local image structures adaptively. The cross-based
filter has some resemblance with the classic bilateral filter, when binarizing the support weight and imposing
a spatial connectivity constraint. For edge-preserving smoothing, our cross-based filter is capable of reaching
similar performance as bilateral filter, while being dozens of times faster. The proposed filter can be applied in
near-constant time, using the integral images technique. In addition, the cross-based filter is highly parallel and
suitable for parallel computing platforms, e.g. GPUs. The strength of the proposed filter is illustrated in several
applications, i.e. denoising and image abstraction.