Image classification algorithms play an important role in various computer vision problems such as object tracking, image labeling, and object segmentation. A number of methodologies have been proposed to tackle this problem. One of the possible approaches employed extensively in the literature is to represent an image as a graph based on its handcrafted features. However, recent advancements in deep neural networks have shown their ability to learn more discriminative and representative features. Therefore, the deep features have become considerable alternatives of hand-crafted ones. In this paper, we propose a novel framework based on distortion-free graph embedding using deep features and KNN-Random forest. Our method outperforms the state-of-the-art graph embedding-based image classification approach for the task of image classification. Particularly, the proposed framework obtains 97.5% top - 1 image classification accuracy for the ImageNet dataset for 5 classes and 93.3% for 10 classes.
|Publication status||Published - May 13 2020|
|Event||17th Conference on Computer and Robot Vision (CRV)|
- Ottawa, ON, Canada, Ottawa, Canada
Duration: May 13 2020 → May 15 2020
|Conference||17th Conference on Computer and Robot Vision (CRV)|
|Period||5/13/20 → 5/15/20|