Applying Transfer Learning to Recognize Clothing Patterns Using a Finger-Mounted Camera

Lee Stearns,Leah Findlater,Jon E. Froehlich
DOI: https://doi.org/10.1145/3234695.3241015
2018-10-08
Abstract:Color identification tools do not identify visual patterns or allow users to quickly inspect multiple locations, which are both important for identifying clothing. We are exploring the use of a finger-based camera that allows users to query clothing colors and patterns by touch. Previously, we demonstrated the feasibility of this approach using a small, highly-controlled dataset and combining two image classification techniques commonly used for object recognition. Here, to improve scalability and robustness, we collect a dataset of fabric images from online sources and apply transfer learning to train an end-to-end deep neural network to recognize visual patterns. This new approach achieves 92% accuracy in a general case and 97% when tuned for images from a finger-mounted camera.
What problem does this paper attempt to address?