Face is a unique characteristic of the human. Detecting the state of the human face, due to its difficulty on the one hand and its many useful features on the other hand, is one of the most important issues in the image processing. In this paper, a five-layer perceptron artificial neural network (MLP) with a supervisor as a complete connection has been used to separate the different facial modes. Learning in the MLP network is done deeply with a high number of layers. The network has 4 class: anger, fear, happiness and surprise. First, the main points and areas of the face that are effective in detecting the state of the face are extracted by edge finding, and then, using the matching of the Fourier series diagram on the operational points of the face, the diagram of those points is obtained. From this diagram, a number of features in the form of three coefficients and an angular velocity are used for network training. Face database images with fixed backgrounds are used for network training. This network is first implemented with Matlab and then MLP layer multiplex is used to implement on FPGA. The results show that the proposed method can be implemented on FPGA platforms with low cost and limited resources, with appropriate output accuracy. In this paper, in addition to speed, accuracy has been tried to create an application system for communication between humans and computers.