Recent studies have shown that deep neural networks (DNNs) are susceptible to adversarial examples attacks, which involve adding imperceptible perturbations to raw images. In a black-box setting, where the attacker lacks knowledge of the model structure and parameters, the success rate of the attack is relatively lower. To tackle this issue, we propose a method to enhance the transferability of adversarial examples by employing Singular Value Decomposition (SVD) to transform the original input image. In our method, the input image is decomposed and reconstructed using SVD in each iteration. After processing with Gaussian kernel, it is combined with the raw image to achieve the transformation. Extensive experimental results validate our method’s effectiveness and demonstrate its potential for easy integration into existing state-of-the-art attack methods in order to significantly improve attack transferability.