In industrial quality control, automated surface inspection is still a major challenge, especially for small, low-contrast, or patterned defects. This paper offers a comparative analysis of contemporary deep-learning detectors, traditional image processing techniques, and a new Hybrid Contour–CNN framework that combines convolutional neural network classification with contour-driven defect localization. Through further acquisition and synthetic augmentation, the ceramic tile defect dataset was increased from 1,750 to 2,550 images, encompassing a variety of surface textures, lighting conditions, and defect types. 5-fold stratified cross-validation was used to benchmark the suggested hybrid model against YOLOv5s, YOLOv8n, SSD-MobileNetV2, a lightweight CNN, and a contour-based pipeline. According to the results, when compared to standalone detectors (YOLOv5s F1 = 70.6% ± 2.8, YOLOv8n F1 = 74.7% ± 1.9), the hybrid model achieves the highest F1 score (88.3% ± 1.5) and improves the detection of subtle defects.