Abstract | ||
---|---|---|
In this paper we explore the role of scale for improved feature learning in convolutional networks. We propose multi-neighborhood convolutional networks, designed to learn image features at different levels of detail. Utilizing nonlinear scale-space models, the proposed multi-neighborhood model can effectively capture fine-scale image characteristics (i.e., appearance) using a small-size neighborhood, while coarse-scale image structures (i.e., shape) are detected through a larger neighborhood. In addition, we introduce a scalable learning method for the proposed multi-neighborhood architecture and show how one can use an already-trained single-scale network to extract image features at multiple levels of detail. The experimental results demonstrate the superior performance of the proposed multi-scale multi-neighborhood models over their single-scale counterparts without an increase in training cost. |
Year | DOI | Venue |
---|---|---|
2015 | 10.1109/MLSP.2015.7324361 | 2015 IEEE 25th International Workshop on Machine Learning for Signal Processing (MLSP) |
Keywords | Field | DocType |
convolutional neural networks,scale-space models,feature extraction,image classification | Nonlinear system,Feature detection (computer vision),Computer science,Artificial intelligence,Computer vision,Architecture,Pattern recognition,Convolution,Feature (computer vision),Feature extraction,Feature learning,Machine learning,Scalability | Conference |
ISSN | Citations | PageRank |
1551-2541 | 0 | 0.34 |
References | Authors | |
18 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Elnaz Barshan | 1 | 0 | 1.01 |
Paul W. Fieguth | 2 | 612 | 54.17 |
Alexander Wong | 3 | 207 | 24.22 |