Currently, mass production of products has become possible owing to the advances in manufacturing technology, enabling the manufacture of products in larger quantities and at higher rates than that done previously. The inspection processes for detecting nonconforming products must be performed at higher speeds without any loss in accuracy. This has motivated the use of online inspection machines such as cameras that detect defective products in many inspection processes. However, accurate criteria must be set for such online inspection machines to distinguish between the non-defective and defective products. In this study, we focus on inspection techniques used to improve the classification accuracy of an actual inspection process. While previous studies have improved the classification accuracy by using a significant amount of preprocessing and feature-extraction calculations, neither to what extent each method affects the accuracy nor how to combine the different methods to improve the accuracy is known. We introduce orthogonal arrays to test the effect of several factors by performing a few experiments. In this study, we construct an optimal combination of methods using the orthogonal array to achieve a higher accuracy and demonstrate that the resulting combination of methods does, achieve a higher classification accuracy than the previous methods.