Carleton University - School of Computer Science Honours Project
Winter 2017
Edge-Aware Image Filtering with Domain Transform and Cumulative Distance
SCS Honours Project Image
ABSTRACT
A new edge-aware image filter is implemented that combines techniques used by two other filters. The first is the domain transform filter, which decomposes the filtering process into horizontal and vertical passes to improve performance by reducing dimensionality. The second is the cumulative range geodesic filter, which includes deviation from the starting pixel colour in its distance formula in order to abstract an image without flattening surfaces or eliminating the impression of texture, resulting in a non-photorealistic painterly effect. The new filter aims to preserve the performance properties of the domain transform method while imitating the texture preservation properties and painterly effect of the cumulative range filter. The filter implementation's performance is good, especially when parallelized using the GPU, but not as good as the original domain transform. Likewise, its texture preservation is hampered by the softening of sharp edges, and it only roughly approximates the painterly effect of the cumulative range filter.