The spatial structure of the surface layer, or texture is important for surface topographycharacterization. In many respects a texture determines contact behavior of the rough surfaces.Despite increasing role of the precision mechanics, the texture of engineering surfaces have not beenadequately investigated. In this paper pattern recognition theory is introduced to perform surfacetextures classification. The height-coded images obtained by atomic force microscopy were used asinitial data. The images represent the surface textures of various materials formed by variousprocesses. We take the following procedure for the texture classification. First, the texture wascharacterized by a matrix of co-occurrence of image contrast. Next, the matrix is transformed intofeature vector by the Karhunen-Loeve transformation. The feature vector was considered ascoordinates of a point in the multidimensional feature space. The location of the point depends on thepeculiarities of the surface texture. The set of the points form clusters that correspond to differentclasses of textures. The mutual arrangement of the points and structure of the clusters were analyzedby the multidimensional scaling procedure. It was founded that there is at least four classes of surfacerelives. The first three of them related to the properties of surface material and the last to the processof growth and crystallization on the interface of different materials.
展开▼
