Individualized Smile Detection Using Transfer Learning
Year: 2017
Type: dissertation
Abstract: In current machine learning approaches to visual smile detection the training data generally\nconsists of a variety of different faces displaying a smile or neutral expression, and the\ntarget data introduces a new face to classify. Although this approach has shown promising\nresults towards correctly classifying the newly introduced target face, we believe that\nperformance can be improved by acknowledging that there are differences in smile-styles\nand facial differences throughout the training process. In this research we aim to do so by\napplying transfer learning. This entails first training a classifier on a data set showing a\nwide variety of smiling and non-smiling faces, as is the general approach. Afterwards we\napply transfer learning, by means of an Adaptive SVM, where a small number of labeled\ninstances from the target face is provided as to make the classifier more specific to the target\nface. To make transfer learning effective we use low-level geometric features which we\nexpect to capture the difference between smiles and no-smile in a variety of different faces\nand smile styles.\nWe evaluate our approach by comparing performance against alternative strategies. We\nfind that using a traditional classifier trained on an aggregated data set containing the general\nand target data outperforms our baseline and our suggested transfer learning approach\nfor each of the test videos. A big problem for the transfer learning approach seems to be\nthe quality of the labeled data of the target face used during training. The aggregated approach\nseems less effected by it, making it a preferred approach for real-life applications\nwhere labeled target data will be sparse and difficult to select. However, its performance\nimprovements does not outweigh its efforts in current experiments, and further research\nshould focus on the choice of target data to use during the training process.
Author: Inge Becht
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2604065181