Return to top English Japanese
Mathematical morphology and texture images
photo

Texture is an image consisting of repeated or random arrangements of simple objects. Reconition and discrimination of textures are important aims of image processing, as well as recognition of object shapes. "Texture modeling" is a texture characteristics description method that considers a model that a texture is an arrangement of grains derived from certain primitives, and that estimates the primitives from the texture under some assumption on the model. This research realizes a simple description of the characteristics of textures, texture descrimination, etc.

We consider a model that a texture is an arrangement of grains derived from one or a few primitives. A texture is characterized using this model by the following three parameters:

  • Shape of primitives,
  • Derivation of grains, and
  • Spatial arrangement of grains (point configuration).

This idea is one of the structural approaches, which describe rules of repetitive structures in textures, and we call this idea "Primitive, Grain and Point Configuration (PGPC)" texture model.

If we introduce an assumption on one of the above three parameters of this model, we can estimate the others. We investigates the estimation of texture characteristics using the mathematical morphology, which is a fundamental theory of shape analysis, and stochastic optimization methods to obtain the most suitable parameters.

Optimization of image processing filters is an application of the texture feature estimation. Applying an image processing filter with parameters fit to the estimated texture features separates noise from the features contained in the target texture and achieves effective noise removal.

Modification of the estimated texture features enables the modification of local and/or global features of the target texture and the generation of a new texture by mixing different features extracted from two different textures. We apply this method to the research of human kansei on images. This research investigates influences of texture features to kansei by generating mixed textures and measuring human kansei on them.

We also organize le Club des Morphologistes Mathématiques du Japon, which is a relationship of researchers who have interests in mathematical morphology in various research area. We organize special sessions on mathematical morphology in various conferences, publish reviews and have review talks, and obtain research budgets, based on le Club.

Selected publications

Texture feature estimation and application to image processing filters

Texture generation

Reviews on mathematical morphology

English Japanese