The use of heterogeneous computing (CPU and GPU) in general purpose application processing has evolved exponentially in recent years. This type of architecture aims to accelerate the achievement of results, which contributes to the reduction in processing time. For this reason, these systems have also been used in applications that need to recognize some type of pattern, such as facial recognition. Facial recognition systems have gained notoriety in recent years because they are more accurate and non-invasive in the process of identifying people previously registered in a database. With this objective, this work uses two parallel libraries, TensorFlow and PyTorch, in order to evaluate the reduction of processing time in a facial recognition application that uses an Artificial Neural Network. The results were promising with the reduction of processing time up to three times when compared to sequential processing time.