.png){kind=icon}
返回
Semantic embedding: scene image classification using scene-specific objects
.png){kind=icon}
PRE
AI
DOI：10.1007/s00530-022-01010-9.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Visual scene understanding is a hot and challenging topic in image processing that aims to understand the general (global) concept of a scene image. In this paper, we propose a novel image embedding algorithm using a learned embedded space, which introduces a high-level semantic representation of the scene images. The learned embedded space as a suitable semantic framework for visual concepts can be used in most applications such as image captioning, Visual Question Answering (VQA), and scene recognition or classification. Inspired by the human inference mechanism in visual scene understanding, the proposed method learns a semantic embedded space of visual concepts using prior semantic knowledge. Prior knowledge is extracted from ConceptNet as one of the most comprehensive knowledge graphs in the form of semantic vectors and is transformed to the learned embedded space with a transformation function. The transformation function is learned by solving a minimization problem. To evaluate our proposed approach, we introduce a scene image dataset called Scene23, which is based on the VisualGenome dataset. A non-linear SVM classifier is utilized to classify the representations of images to the scene categories. The experimental results show 99.44% classification accuracy on the Scene23 dataset. Also, we evaluated our proposed method by the UIUC Sports and MIT67 datasets. Experimental results indicate that our proposed method outperforms other state-of-the-art methods on the UIUC Sports dataset and achieves competitive results on the MIT67 dataset.
Keyword:
Scene classification
Semantic embedding
Scene-specific objects
期刊
IF:
3.1
论文数:
2.8K
被引数:
2.7K
机构
引用论文
分享收藏Indoor scene modeling from a single image using normal inference and edge features使用法线推理和边缘特征从单个图像进行室内场景建模.png)
VISUAL COMPUTER
VISUAL COMPUTERIF2.9
PREAI
分享收藏MICAL2is a novel human cancer gene controlling mesenchymal to epithelial transition involved in cancer growth and invasionOncotarget
OAAI
OncotargetIF0
OAAI分享收藏
查看更多内容