This work presents a shallow network based on subspaces with applications in imageclassification. Recently, shallow networks based on PCA filter banks have beenemployed to solve many computer vision-related problems including textureclassification, face recognition, and scene understanding. These approaches are robust,with a straightforward implementation that enables fast prototyping of practicalapplications. However, these architectures employ either unsupervised or supervisedlearning. As a result, they may not achieve highly discriminative features in morecomplicated computer vision problems containing variations in camera motion,object’s appearance, pose, scale, and texture, due to drawbacks related to eachlearning paradigm. To cope with this disadvantage, we propose a semi-supervisedshallow network equipped with both unsupervised and supervised filter banks,presenting representative and discriminative abilities. Besides, the introducedarchitecture is flexible, performing favorably on different applications whose amount ofsupervised data is an issue, making it an attractive choice in practice. The proposednetwork is evaluated on five datasets. The results show improvement in terms ofprediction rate, comparing to current shallow networks.
展开▼