Texture Features Extraction Using Wavelet-Based CLBP for Surface Defects Retrieval of Hot Rolled Steel Strip

This paper addresses the effective content-based image retrieval (CBIR) system specifically designed for identifying surface defects of hot-rolled steel strips. This system allows the expert to separate serious defects from others automatically. The proposed CBIR system is based on extracting completed local binary patterns (CLBP) features from the discrete wavelet transform (DWT) coefficients. This DWCLBP descriptor extracts highly discriminative texture features at different scales and orientations. We select the optimal mother wavelet and wavelet decomposition levels by making extensive comparisons. The experimental results on NEU surface defects database demonstrate promising retrieval performance of the DWCLBP features.