Facial expression and emotions recognition had attracted the attention of game developers and there are some game proposals, in which, by detecting faces, it is possible to detect who is playing and automatically load the players' profile. Likewise, the recognition of the players' facial expressions has been used to alter the expression of their avatars. In this work, we intended to study ways to use the recognition of a player's facial expression to change some parameters of a game, such as stimulate it when the player is bored or increasing its difficulty, for example, when it detects that the player is more stimulated and happy. This paper proposes a new approach for facial expression and emotion recognition, which divides the classification stage into two steps. First it is detected if it is a positive, a negative or a neutral emotion. Then it is applied a refinement step to recognize the particular positive (happiness or surprise) or negative (sadness, fear, anger or disgust) emotion. Our approach runs in real-time and was compared with the most common classification methods and has shown a great accuracy. We built a game to test our approach. The estimation of the player's emotional state was used to change certain parameters of the game, which increases or decreases the game difficulty.