随着5G和网络技术的飞速发展,大量互联网图像出现在人们的视野中。互联网图像的高维和噪声特性是图像分类问题的主要挑战。为提高互联网图像的识别性和鲁棒性,本文提出了一种鲁棒多视角潜在低秩表示(robust multi-view latent low rank representation, RMLLRR)的图像分类方法。RMLLRR算法在低秩表示算法的框架上引入多视角学习的思想,根据视角互补性和一致性准则,利用多种特征得到图像全面的描述信息,最大化不同视角间的一致性和最小化视角间信息描述的分歧。RMLLRR算法使用潜在低秩表示的思想,过滤冗余特征和噪声信息,着重考虑图像主要特征信息和显著特征信息,使得模型更加鲁棒和分辨力。此外,RMLLRR算法运用ε-draggings技术学习类间大间隔的松弛标签矩阵,起到增强类别判别的作用。人脸数据集ORL、物体数据集COIL和对象识别数据集GRAZ的实验结果表明,在噪声环境下,RMLLRR算法在所有对比算法中取得了最好的分类结果,分类精度分别达到92.43%、98.95%和63.37%。
Abstract
With the rapid development of 5G and network technology, a large number of internet images have appeared in people′s vision. The high-dimension and noise characteristics of Internet images are the main challenges in classification problems. In order to improve the recognition and robustness of Internet images, this study proposes a robust multi-view latent low rank representation (RMLLRR) algorithm for image classification. The RMLLRR algorithm incorporates the idea of multi-view learning within the framework of low rank representation algorithm. Based on the complementary and consistent criteria of multiple views, it utilizes multiple features to obtain comprehensive image description information, maximizing consistency between different views and minimizing divergence in information description between views. The RMLLRR algorithm uses the idea of latent low rank representation, filters redundant features and noise information, and focuses on the principal feature and salient feature of the image, making the model more robust and discriminative. In addition, the RMLLRR algorithm utilizes the ε-draggings technique to learn the relaxed label matrix with large intervals between classes, which enhances the discrimination ability of classes. The experimental results of the face dataset ORL, object dataset COIL, and object recognition dataset GRAZ show that in noisy environments, the RMLLRR algorithm achieves the best classification results among all compared algorithms, with classification accuracy of 92.43%, 98.95%, and 63.37%, respectively.
关键词
多视角学习 /
潜在低秩表示 /
ε-draggings技术 /
图像分类
{{custom_keyword}} /
Key words
multi-view learning /
latent low-rank representation /
ε-draggings technology /
image classification
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] HAO Y, JING X, CHEN R, et al. Learning enhanced specific representations for multi-view feature learning[J]. Knowledge-Based Systems, 2023, 272(7): 110590.
[2] TANG J, LI D, TIAN Y, Image classification with multi-view multi-instance metric learning[J]. Expert Systems with Applications, 2023, 189(3): 116117.
[3] WANG Z, SHEN Z, ZOU H, et al. Retargeted multi-view classification via structured sparse learning[J]. Signal Processing, 2022, 197(8): 108538.
[4] LIU P, ZHAO S, ZOU Q, et al. Prediction of cell-penetrating peptides using a novel HSIC-based multiview TSK fuzzy system[J]. Applied Sciences, 2022, 12(11): 5383.
[5] 徐光生, 王士同. 基于潜在的低秩约束的不完整模态迁移学习[J]. 计算机科学与探索, 2021, 16(2): 2775-2788.
XU G S, WANG S T. Incomplete modality transfer learning via latent low-rank constraint[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(12): 2775-2787.
[6] CHEN Z, WU X, KITTLER J. Low-rank discriminative least squares regression for image classification[J]. Signal Processing, 2020, 173 (8): 107485.
[7] GAO M, LIU R, MAO J. Noise robustness low-rank learning algorithm for EEG signal classification[J]. Frontiers in Neuroscience, 2021, 24(11): 797378.
[8] LIU Z, SONG P. Deep low-rank tensor embedding for multi-view subspace clustering[J]. Expert Systems with Applications, 2024, 237: 121518.
[9] ZHANG G, HUANG D, WANG C. Facilitated low-rank multi-view subspace clustering[J]. Knowledge-Based Systems, 2023, 260(2): 110141.
[10] WANG J, TIAN F, YU H, et al. Diverse non-negative matrix factorization for multiview data representation[J]. IEEE Trans. Cybernetics, 2018, 48(9): 2620-2632.
[11] LIU G, LIN Z, YAN S, et al. Robust recovery of subspace structures by low-rank representation[J]. IEEE Trans. Pattern Analysis and Machine Intelligence, 2012, 35 (1): 171-184.
[12] SAKTHIVEL S, LAKSHMIPATHI R. Enhancing face recognition using improved dimensionality reduction and feature extraction algorithms-an evaluation with ORL database[J]. International Journal of Engineering Science and Technology, 2010, 2(6):2288-2295.
[13] MODEL I, SHAMIR L. Comparison of data set bias in object recognition benchmarks[J]. IEEE Access, 2015, 3(1):1953-1962.
[14] HEMANDEZ D, CLEMENTE E, OLAGUE G, et al. Evolutionary multi-objective visual cortex for object classification in natural images[J]. Journal of Computational Science, 2016, 1(17):216-233.
[15] VU T, MONGA V. Fast low-rank shared dictionary learning for image classification[J]. IEEE Trans. Image Processing, 2017, 26(11): 5160-5175.
[16] HUANG C, CHUNG F, WANG S. Multi-view L2-SVM and its multi-view core vector machine[J]. Neural Networks, 2016, 75(3): 110-125.
[17] WANG Z, ABHADIOMHEN S E, LIU Z, et al. Multi-view intrinsic low-rank representation for robust face recognition and clustering[J]. IET Image Processing, 2021, 15(14): 3573-3584.
[18] ZHOU J, SHEN X, LIU S, et al. Multi-dictionary induced low-rank representation with multi-manifold regularization[J]. Applied Intelligence, 2023, 53(3): 3576-3593.
[19] ZHANG Y, GAVNANOVA I. Joint association and classification analysis of multi-view data, Biometric Methodology[J]. 2022, 78(4): 1614-1625.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
江苏省高职院校教师专业带头人高端研修项目(2023GRFX004)
{{custom_fund}}
PDF(11279 KB)
