Dog Breed Library with Picture-Based Search Using Neural Networks

In the past decade, there have been significant improvements in image classification thanks to approaches using neural networks. With the widespread availability of free and open-source frameworks for training neural networks, such as Google's TensorFlow, it is now feasible to make use of advanced image classification in novel areas. One such area is the problem of identifying dog breeds from photos. The American Kennel Club currently recognizes 193 distinct breeds-all of which not only look different and have different temperaments, but which may require special care, such as frequent grooming, a specific diet, or a more active lifestyle. Therefore, it is important for dog owners to know their dogs' breeds. The main subject of this paper is the usage of neural networks to provide non-technical users with the means to automatically identify breeds of dogs from their pictures.