Personal identity authentication is the key to ensure personal information security, and it is widely used in banking, security inspection and other fields. However, due to the progress of technology, the security of some personal identity authentication methods has weakened in the past, and incidents such as forged ID cards frequently occur. The security of mobile phone dynamic authentication is also slowly declining. In order to avoid the above problems, face recognition technology came into being, and people paid extensive attention to it. In this paper, the development trend of CNN(Convolutional Neural Network) face recognition technology has been deeply studied and optimized. The research shows that the recognition rate is as high as 90% under normal light, left light and right light, and more than 80% under low light and strong light, which shows that the face recognition in this paper is robust to light changes. The main reasons for the failure of face recognition in low light and strong light are the unclear position of face feature points in the image and the poor quality of the collected image. The quality of the collected image can be judged in the collection stage to improve the recognition rate of this algorithm. The image quality evaluation and convolution kernel design optimization proposed in this study not only improve the accuracy of face recognition technology, but also enhance the stability of the system under different lighting conditions. Compared with the existing face recognition technology based on CNN, this method has achieved remarkable progress in technology through refined image quality screening and targeted convolution kernel design.