Two principal paradigms dominate in automated defect detection: convolutional neural network (CNN)-based approaches and proprietary non-CNN algorithmic methods. The optimal choice depends on the target material characteristics and operational constraints. The current state-of-the-art model on the MVTec Anomaly Detection (AD) Dataset, Global and Local Anomaly co-Synthesis Strategy, employs a CNN architecture and achieves an area under the receiver operating characteristic curve (AUC) of 1.000 on the tile subset and 0.999 on the wood subset. However, practical deployment challenges, including high computational requirements and limited interpretability, remain. Conversely, a non-CNN defect discrimination method based on luminance standard deviation offers low computational cost and high interpretability. Nevertheless, it performs poorly when defects involve subtle color variations or when surfaces exhibit irregular patterns, achieving an area under the AUC of only 0.998 and 0.967, respectively. To address the limitations of conventional single-feature methods, we propose a novel discrimination framework for materials characterized by complex background patterns and diverse defect types. The proposed framework integrates material-specific preprocessing with multiple discrimination methods tailored to the characteristic features of individual defect types, thereby enabling highly accurate defect detection. The proposed framework achieved perfect performance (1.000) across all evaluation metrics—average accuracy, precision, recall, specificity, and AUC—on tile and wood datasets. These results confirm the effectiveness of the proposed framework for defect detection in tile and wood products. Further, the framework is expected to maintain low computational cost and short processing time. However, misclassification risk may increase when surface patterns are highly irregular or when the red, green, and blue variance, luminance values, or luminance variance of defective regions closely resemble those of defect-free regions.