Abstract:
Since the ImageNet competition was held in 2012, the machine learning algorithm has performed very well on the image classification task. In object classification task, t...Show MoreMetadata
Abstract:
Since the ImageNet competition was held in 2012, the machine learning algorithm has performed very well on the image classification task. In object classification task, there are still some problems. For example, similar categories are difficult to be distinguished in images. In addition, with the increasing number of object categories, the background of scene has become another crucial issue in object classification. This paper focuses on image object recognition and makes two major contributions to tackle these issues. Firstly, we propose a semantic refinement method that analyzes the relationship among similar categories in images from the perspective of semantic knowledge. The knowledge of the relationship between semantics comes from a wide range of open knowledge. Secondly, we utilize the knowledge graph method to create the image knowledge graph with multiple categories in images. Our method exploits the knowledge from the adjacency matrix computed on train data to merge relevant classes into graph. We conduct extensive experiments on large-scale image datasets (ImageNet), demonstrating the effectiveness of our approach. Further, our method participates in ILSVRC 2012 challenges, and obtain the new state-of-the-art results on the ImageNet (82.43%).
Published in: 2019 IEEE 23rd International Conference on Computer Supported Cooperative Work in Design (CSCWD)
Date of Conference: 06-08 May 2019
Date Added to IEEE Xplore: 08 August 2019
ISBN Information: