With the advent of the 6G and the era of massive intelligent connectivity, the demand for image transmission has surged dramatically. However, traditional image transmission methods face significant challenges. Ensuring complete image transmission requires substantial communication resources, while a large amount of transmitted data contains task-irrelevant redundancy, leading to inefficiencies. To address these issues, semantic communication has been gaining increasing attention as an emerging communication paradigm, which enhances communication efficiency by extracting and transmitting essential information. However, ensuring reliability in semantic communication remains a critical challenge, as transmission errors can degrade semantic understanding and overall system performance. Unlike conventional communication systems that rely on explicit channel coding for error detection and correction, semantic communication lacks a well-defined error control mechanism. To bridge this gap, we propose a semantic error detector and a semantic error corrector based on a Generative Adversarial Network (GAN) to identify and rectify transmission errors that effectively integrate Hybrid Automatic Repeat reQuest (HARQ) into image semantic communication. Furthermore, we introduce a feature-level fusion method, thereby improving image quality. Experimental results validate the effectiveness of the proposed modules and demonstrate the superiority of semantic HARQ schemes in low-SNR environments.