An Approach to Rotation Invariant Texture Classification and Some Experimental Results.
PURDUE UNIV LAFAYETTE IN SCHOOL OF ELECTRICAL ENGINEERING
Pagination or Media Count:
This paper presents a new feature extraction method for classifying a texture image into one of the n possible classes, C sub i, i 1,...,n. The extracted features are invariant under rotation or gray scale changes. Two types of random field models namely, Circular Auto Regressive model, and Simultaneous Auto Regressive model are used to extract these features. These models are fitted to a given MxM digitized image and their parameters are estimated. These estimated parameters and some functions of them constitute the desired rotation invariant feature vector. The classification power of this feature vector is demonstrated through experimental results obtained with twelve different classes of natural textures including both macrotextures and microtextures. Author
- Theoretical Mathematics