Restoration of old images acquired from repositories of historical photographic and film archive is of key importance for the preservation, presentation and transmission of the legacy of past cultures to the coming generations. These old images are often subject to severe blending degradation, and most related methods are less effective in restoring larger structural degradations like vertical stripes. To better restore multiple mixed degradations in old images, we collect thousands of real old images and synthetic degradation images to build our dataset. Besides, a novel Mixed Feature Attention (MFA) module, which includes channel attention, spatial attention and pixel attention mechanism, is proposed to optimize the restoration of structural degradations and ensure that the restored content of structural degradations can match the surrounding pixels. And we creatively integrate the MFA into a triple domain translation network. The experimental results show that our method can achieve the most advanced performance in terms of quantitative, qualitative comparison and user study, and has more real and natural repair result specially for structural degradation.