Multivariate statistical projection methods to perform robust feature extraction and classification in surface grading

bstract. We present an innovative way to simultaneously perform eature extraction and classification for the quality-control issue of urface grading by applying two multivariate statistical projection ethods: SIMCA and PLS-DA. These tools have been applied to ompress the color texture data that describe the visual appearance f surfaces (soft color texture descriptors) and to directly perform lassification using statistics and predictions from the projection odels. Experiments have been carried out using an extensive ceamic images database (VxC TSG) comprised of 14 different modls, 42 surface classes, and 960 pieces. A factorial experimental esign evaluated all the combinations of several factors affecting he accuracy rate. These factors include the tile model, color repreentation scheme (CIE Lab, CIE Luv, and RGB), and compression/ lassification approach (SIMCA and PLS-DA). Moreover, a logistic egression model is fitted from the experiments to compute accuacy estimates and study the effect of the factors on the accuracy ate. Results show that PLS-DA performs better than SIMCA, chieving a mean accuracy rate of 98.95%. These results outperorm those obtained in a previous work where the soft color texture escriptors in combination with the CIE Lab color space and the -NN classifier achieved an accuracy rate of 97.36%. © 2008 SPIE nd IS&T. DOI: 10.1117/1.2957886