Edge-based texture granularity detection

Directly connected to the texture appearance, texture granularity is an effective measurement for geographic resources classification, product quality monitoring and image compression ratio selection. However, the application of existing works on texture granularity is limited by intense computation and the dependence on empirically selected parameters that vary among different textures. This paper proposes an edge-based texture granularity detection algorithm that takes textures as homogeneous cells separated by prominent boundaries. Experiments on two datasets show that the proposed method yields granularity consistent with perceptual measures and is highly computationally efficient compared to existing methods.