Feature Extraction and Classification from Texture Image of Machined Surfaces Using Multilevel Wavelet Decomposition and Logistic Regression

This manuscript aims to identify the texture images of the different machined operation, with the help of machine vision and artificial intelligence techniques. Texture of machined component, viz. electric discharge machining, milling, sand blasting and shaping is captured and segmented into sixteen equal, non-overlapping sub images and then discrete wavelet transform using Daubechies wavelet is applied on the sub images. Wavelet coefficients of sub images are decomposed up to fourth level, and five significant features are extracted from each level. Logistic regression, which is an artificial intelligent technique, is applied to get training and testing efficiency for identifying texture images. It is observed that the decomposition level 1 gives the 100% training identification as well as 92.3% testing identification. Results revealed that as level of decomposition is increased efficiency decreases and a number of incorrect classified instances rise drastically.