When considering individuals with dietary limitations, automatic food recognition, and assessment are paramount. Smartphone-oriented applications are convenient and handy when dish recognition and the elements inside are required. Machine learning (deep learning) applied to image recognition, alongside other classification techniques (for example, bag-of-words), are possible approaches to tackle this problem. The current most promising approach to the classification problem is deep leaning, which requires high computation for training, but it is an extremely fast and computationally light classifier. Since the requirement for the classifiers to be as accurate as possible, the humans must also be considered as the classifier. This work tests and compares deep-learning methods bag-of-words applied to computer vision, and the human visual system. Results show that deep learning is better when considering a low number of food categories. However, with more food categories, the human overcomes the machine algorithms.