Identity recognition using the biometrical characteristics of facial pixel information extracted from the person has been undoubtedly an important AI technique issue in security and surveillance applications. In identity recognition using facial RGB-sensing pixels, an undesired situation in which classification performance would be degraded is to encounter a blurred face image composed of substandard facial pixels hybridized with blur noise. The generative deep learning network, deblurring generative adversarial network (DeblurGAN), is well known to be effective in removing blur noise from the blurred image to help object detection. Such object detection applications that are focused on the issue of public safety accomplish person detection by using detection networks (such as YOLO) in fast flows of walking or running pedestrians. DeblurGAN in person detection enhancements mainly improves only the images with motion blur. Another typical type of substandard image that is also frequently used in surveillance applications with the high protection of personal privacy is an image with Gaussian blur. However, studies on exploring the effectiveness of blur alleviation of Gaussian-blurred images by DeblurGAN as well as evaluating the performance of visual geometry group network (VGGNet) identity classification are rare. To tackle this issue, in this study, DeblurGAN is used to alleviate blurring in face images with motion or Gaussian blur, followed by the evaluation of the recognition performances of VGGNet identity classifications using two different types of substandard facial pixels with blurring. A series of performance analysis and comparison experiments are carried out using designed face image datasets composed of sharply focused faces without blur disturbance, motion-blurred and Gaussian-blurred faces with different degrees of blur, and DeblurGAN-restored faces with blur alleviation in the VGGNet identity recognition task. Various important points are constructively identified in this study on the basis of the recognition performance results observed in identity classification experiments, which will provide a fine reference for the development of practical application systems in real life by VGGNet identity classification incorporated with the online blur detection and DeblurGAN blur improvement of the blurred face image.