For analyzing and designing the face recognition system. Difference methods can be used in designing and training the model are such like using face landmark by 68 points or 468 points. In my research is not going deeper with the face landmark research to achieve a face recognition system. This is because using face landmark method in training a model will take long time in training and easily occur error during deep network learning process. And, using face landmark method on real time face recognizing is easily conducted long time on computing process for recognizing facial and easily conducted distortion when multi facial occur at once time. Therefore, my research is toward another method which is using architecture for training a model. The benefit of using architecture or network on training a model is this method able to let user set or modify their custom layer on architecture used. In my research, the face landmark is just for detecting purpose for input providing and recognizing process is carrying by architecture layer matching. And, my recognizing process is based on high similarity matching with database image which this method is consume less time in computing and fast response than fully depending face landmark system. 5 architecture networks have been researched and trained in my research model are ResNet50, ResNet152, MobileNet, VGG16 and Xception. The reason for doing difference architectures researched and trained is trying to compare the ability, accuracy and data result in training time, training loss, validation loss, training accuracy and validation accuracy. Lastly, providing the most suitable method, architecture used and model trained for planning an automatic door lock system.