Food recognition on smartphone using transfer learning of convolution neural network

Food recognition is one challenging domain on computer vision because of the complex structure of food images. On the other hand, it is a worthy issue because of its versatile applications e.g. monitoring dietary consumption of aging people and patients or finding defects in food processing line. In this paper, we propose a new pipeline to recognize a set of Thai food images based on transfer learning technique of convolution neural network. Learning steps and image distortions were two primary experimented parameters. Testing on 40 categories of totally 1987 images revealed that the proposed pipeline gave a promising accuracy at 75.2%.