There are many important applications for anomaly detection, such as industrial image anomaly detection, medical image anomaly detection etc. The teacher-student network (T-S) offers an effective method for identifying anomalies. Previous studies mainly utilize normal data or pseudo-anomalous for training, relying on cosine similarity for differentiation and usually involving only one guiding teacher. This approach limits the learning capacity of the student model, and solely concentrating on pixel-level differences might may hinder the model's grasp of overall anomaly structure and context, resulting in inadequate abnormal behavior recognition. To solve the problems, Dual-Teacher Network with SSIM based on Reverse Distillation (DSRD) is proposed, which combines the functionalities of two pre-trained teacher networks, namely T1 and T2, along with a denoising student encoder-decoder (S), a trainable one-class bottleneck embedding module (OCBE), and a segmentation network, all integrated within a unified framework. This comprehensive approach aims to leverage the strengths of each component to enhance outlier point detection and anomaly recognition capabilities. In the experiments on industrial inspection benchmark datasets, our method achieved state-of-the-art performance. The image-level AUC reached 99.41%, the pixel-level AUC reached 98.80%, the pixel-level average precision reached 79.34%, and the AUPRO reached 94.91%.