Turkish cuisine: A benchmark dataset with Turkish meals for food recognition

Food recognition in still images is a problem that has been recently introduced in computer vision. The benchmark data sets used in training and evaluation of food recognition methods contain sample images of popular foods from the globe. However, when they are examined thoroughly, it can be observed that very few of them are Turkish dishes. In this study, we first carry out a data collection process for Turkish dishes and construct a new dataset named "TurkishFoods-15" containing 500 images in each food class. In addition, we introduce a novel food recognition approach that depends on fine-tuning Google Inception v3 deep neural network model based on transfer learning. For this purpose, our Turkish cuisine dataset was combined with the widely used Food-101 dataset from the literature and the performance analysis of the developed deep learning-based approach is carried out on this combined dataset containing 113 food classes. Our results show that the recognition of Turkish dishes can be achieved with certain success even though it does not have certain difficulties.