Pixel Classification By Morphologically Derived Texture Features

摘要: Local granulometric size distributions are generated by performing a granulometry on an image and keeping local pixel counts in neighborhood of each at the completion successive opening. Normalization resulting yields probability density pixel. These densities contain texture information to Pixels can be classified according moments densities. Further refinement accomplished employing several structuring-element sequences order generate number granulometries, revealing different qualities. Classification is comparing observed those representing database textures. The collection actually random variables dependent processes, method employed present paper involves comparison means database-texture moments.