Image analysis by information decomposition and synthesis

The authors compare the image analysis capability using the linear Gaussian pyramid and nonlinear morphological pyramid image decomposition/synthesis techniques, with emphasis on image enhancement. In the experiments, the authors used the weakening strategy of the `between band correlation then synthesis´ approach to remove two types of noise: the unstructured noise (e.g., the random sparkle noise) and the structured noise (e.g., the high-frequency imaging artifact) from an image which has been decomposed either by morphological pyramid or Gaussian pyramid techniques. White sparkle noise was successfully removed by the morphological pyramid band decomposition/synthesis approach based on an opening operation with a flat 3×3 cross structure element. In contrast, the Gaussian pyramid approach with 5×5 Gaussian kernel was not entirely successful in removing the same noise. A similar result was obtained when using both approaches for removing the structured noise