Data-Independent Feature Learning with Markov Random Fields in Convolutional Neural Networks

被引:3
作者
Peng, Yao [1 ]
Hankins, Richard [1 ]
Yin, Hujun [1 ]
机构
[1] Univ Manchester, Sch Elect & Elect Engn, Manchester M13 9PL, Lancs, England
关键词
Convolutional neural networks; Image representation; Markov random fields; Gibbs distribution; Self-organising maps; Image classification; Image features; SPATIAL-INTERACTION; CONVERGENCE; MODELS; SPACE;
D O I
10.1016/j.neucom.2019.03.107
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
In image classification, deriving robust image representations is a key process that determines the performance of vision systems. Numerous image features and descriptors have been developed manually over the years. As an alternative, however, deep neural networks, in particular convolutional neural networks (CNNs), have become popular for learning image features or representations from data and have demonstrated remarkable performance in many real-world applications. But CNNs often require huge amount of labelled data, which may be prohibitive in many applications, as well as long training times. This paper considers an alternative, data-independent means of obtaining features for CNNs. The proposed framework makes use of the Markov random field (MRF) and self-organising map (SOM) to generate basic features and model both intra- and inter-image dependencies. Various MRF textures are synthesized first, and are then clustered by a convolutional translation-invariant SOM, to form generic image features. These features can be directly applied as early convolutional filters of the CNN, leading to a new way of deriving effective features for image classification. The MRF framework also offers a theoretical and transparent way to examine and determine the influence of image features on performance of CNNs. Comprehensive experiments on the MNIST, rotated MNIST, CIFAR-10 and CIFAR-100 datasets were conducted with results outperforming most state-of-the-art models of similar complexity. (C) 2019 Elsevier B.V. All rights reserved.
引用
收藏
页码:24 / 35
页数:12
相关论文
共 50 条
[41]   PSO-Convolutional Neural Networks With Heterogeneous Learning Rate [J].
Phong, Nguyen Huu ;
Santos, Augusto ;
Ribeiro, Bernardete .
IEEE ACCESS, 2022, 10 :89970-89988
[42]   Markov Random Fields for Malware Propagation: The Case of Chain Networks [J].
Karyotis, Vasileios .
IEEE COMMUNICATIONS LETTERS, 2010, 14 (09) :875-877
[43]   Feature extraction and classification of VHR images with attribute profiles and convolutional neural networks [J].
Tian Tian ;
Lang Gao ;
Weijing Song ;
Kim-Kwang Raymond Choo ;
Jijun He .
Multimedia Tools and Applications, 2018, 77 :18637-18656
[44]   Remote sensing object detection with feature-associated convolutional neural networks [J].
Rao, Jianghao ;
Wu, Tao ;
Li, Hongyun ;
Zhang, Jianlin ;
Bao, Qiliang ;
Peng, Zhenming .
FRONTIERS IN EARTH SCIENCE, 2024, 12
[45]   Inductive conformal predictor for convolutional neural networks: Applications to active learning for image classification [J].
Matiz, Sergio ;
Barner, Kenneth E. .
PATTERN RECOGNITION, 2019, 90 :172-182
[46]   Smart feature extraction and classification of hyperspectral images based on convolutional neural networks [J].
Hamouda, Maissa ;
Ettabaa, Karim Saheb ;
Bouhlel, Med Salim .
IET IMAGE PROCESSING, 2020, 14 (10) :1999-2005
[47]   Feature-Based Interpretation of Image Classification With the Use of Convolutional Neural Networks [J].
Wang, Dan ;
Xia, Yuze ;
Yu, Zhenhua .
IEEE ACCESS, 2024, 12 :70377-70391
[48]   Intraday prediction of Borsa Istanbul using convolutional neural networks and feature correlations [J].
Gunduz, Hakan ;
Yaslan, Yusuf ;
Cataltepe, Zehra .
KNOWLEDGE-BASED SYSTEMS, 2017, 137 :138-148
[49]   Feature extraction and classification of VHR images with attribute profiles and convolutional neural networks [J].
Tian, Tian ;
Gao, Lang ;
Song, Weijing ;
Choo, Kim-Kwang Raymond ;
He, Jijun .
MULTIMEDIA TOOLS AND APPLICATIONS, 2018, 77 (14) :18637-18656
[50]   Spatial-spectral classification of hyperspectral images: a deep learning framework with Markov Random fields based modelling [J].
Qing, Chunmei ;
Ruan, Jiawei ;
Xu, Xiangmin ;
Ren, Jinchang ;
Zabalza, Jaime .
IET IMAGE PROCESSING, 2019, 13 (02) :235-245