Study of distance metrics for surface metrology application with dual-tree complex wavelet transform

Surface Metrology deals with study of surface textures. The dual-tree complex wavelet transform (DT-CWT) is used for analyzing texture images of engineered surfaces. The metrological characteristics of DT-CWT motivated us to use it as a tool for analysis. We use energy, standard deviation and weighted standard deviation (WSD) as texture descriptors for surface texture classification. The feature set includes 24 features. Eight different distance metrics are tested for their performance with three surface texture databases namely Milling, Casting and Shaping. The class of the database is designated by its roughness value. It is found that Squared Chi-square, Squared Chord, Canberra and Bray-Curtis distance measures perform well for classification of engineered surfaces. We achieved the overall classification performance of 84.55%. We compared the classification results of DT-CWT and real discrete wavelet transform (DWT). It shows the improvement in the correct rate of classification from 80% to 88.33% on Milling database, from 62.22% to 74.07% on Casting database and from 82.92% to 91.25% on Shaping Database. The class associated with the test image is the roughness of the surface.