اصلاح معماری شبکه عصبی کانولوشنال جهت طبقه‌بندی تصاویر آغشته به نویز ضربه
محورهای موضوعی : مهندسی برق و کامپیوتر
محمد مومنی
1
,
مهدی آقاصرام
2
,
علی محمد لطیف
3
,
راضیه شیخ پور
4
1 - دانشگاه یزد
2 - دانشگاه یزد
3 - دانشگاه یزد
4 - دانشگاه اردكان
کلید واژه: نویز ضربهشبکه عصبی کانولوشنالطبقه‌بندی تصویرشناسایی نویز,
چکیده مقاله :
نویز ضربه موجب اختلال در فرایند طبقه‌بندی تصاویر توسط شبکه عصبی کانولوشنال می‌گردد. پیش‌پردازش جهت حذف نویز ضربه هزینه‌بر است و تصاویر تخریب‌شده به دلیل عدم بهبود کافی، اثر مخرب در مراحل آموزش و اعتبارسنجی این شبکه دارند. در این مقاله با اصلاح معماری شبکه عصبی کانولوشنال، یک مدل مقاوم در برابر نویز ضربه معرفی می‌شود. روش پیشنهادی، طبقه‌بندی تصاویر نویزی را بدون نیاز به هیچ گونه پیش‌پردازش انجام می‌دهد. لایه تشخیص نویز ضربه در بدنه شبکه عصبی کانولوشنال تعبیه می‌شود و از پردازش مقادیر نویزی جلوگیری می‌کند. برای آموزش مدل پیشنهادی از پایگاه داده 2012- ILSVRC استفاده شده است. نتایج شبیه‌سازی نشان می‌دهد که جلوگیری از تأثیرگذاری نویز ضربه در فرایند آموزش و طبقه‌بندی شبکه عصبی کانولوشنال، دقت و سرعت آموزش شبکه را افزایش می‌دهد. روش پیشنهادی با خطای 24/0 در طبقه‌بندی تصاویر آغشته به نویز ضربه با چگالی 10% بهتر از سایر روش‌های مورد مقایسه می‌باشد. مرتبه زمانی (1)O در اصلاح CNN جهت مقاومت در برابر نویز نشان‌دهنده برتری روش پیشنهادی است.
Impulse noise is one the common noises which reduces the performance of convolutional neural networks (CNNs) in image classification. Preprocessing for removal of impulse noise is a costly process which may have a destructive effect on the training and validation of the convolutional neural networks due to insufficient improvement of noisy images. In this paper, a convolutional neural network is proposed which is robust to impulse noise. Proposed CNN classify images corrupted by impulse noise without any preprocessing for noise removal. A noise detection layer is placed at the beginning of the proposed CNN to prevent the processing of noisy values. The ILSVRC-2012 database is used to train the proposed CNN. Experimental results show that preventing the impact of impulse noise on the training process and classification of CNN can increase the accuracy and speed of the network training. The proposed CNN with error of 0.24 is better than other methods in classification of noisy image corrupted by impulse noise with 10% density. The time complexity of O(1) in the proposed CNN for robustness to noise indicates the superiority of the proposed CNN.
[1] Z. Zhang, D. Han, J. Dezert, and Y. Yang, "A new adaptive switching median filter for impulse noise reduction with pre-detection based on evidential reasoning," Signal Processing, vol. 147, pp. 173-189, Jun. 2018.
[2] K. H. Jin and J. C. Ye, "Sparse and low-rank decomposition of a hankel structured matrix for impulse noise removal," IEEE Trans. Image Process., vol. 27, no. 3, pp. 1448-1461, Nov. 2018.
[3] I. Turkmen, "The ANN based detector to remove random-valued impulse noise in images," J. Vis. Commun. Image Represent., vol. 34, pp. 28-36, Jan. 2016.
[4] S. Liang, S. Lu, J. Chang, and C. C. T. Lin, "A novel two-stage impulse noise removal technique based on neural networks and fuzzy decision," IEEE Trans. on Fuzzy Systems, vol. 16, no. 4, pp. 863-873, Aug. 2008.
[5] Y. Hou, Z. Li, P. Wang, and W. Li, "Skeleton optical spectra-based action recognition using convolutional neural networks," IEEE Trans. Circuits Syst. Video Technol., vol. 28, no. 3, pp. 807-811, Nov. 2018.
[6] C. Ding and D. Tao, "Trunk-branch ensemble convolutional neural networks for video-based face recognition," IEEE Trans. Pattern Anal. Mach. Intell., vol. 40, no. 4, pp. 1002-1014, May 2018.
[7] L. C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, "DeepLab: semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs," IEEE Trans. Pattern Anal. Mach. Intell., vol. 40, no. 4, pp. 834-848, Apr. 2018.
[8] G. Lin, Q. Wu, L. Qiu, and X. Huang, "Image super-resolution using a dilated convolutional neural network," Neurocomputing, vol. 275, pp. 1219-1230, Jan. 2018.
[9] S. Yu, S. Jia, C. Xu, "Convolutional neural networks for hyperspectral image classification," Neurocomputing, vol. 219, pp. 88-98, Jan. 2016.
[10] K. Audhkhasi, O. Osoba, B. Kosko, "Noise-enhanced convolutional neural networks," Neural Networks, vol.78, pp. 15-23, Jun. 2016.
[11] W. Zhang, L. Jin, E. Song, and X. Xu, "Removal of impulse noise in color images based on convolutional neural network," Applied Soft Computing, vol. 82, Article No. 105558, Sept. 2019.
[12] L. Jin, W. Zhang, G. Ma, and N. Song, "Learning deep CNNs for impulse noise removal in images," Journal of Visual Communication and Image Representation, vol. 62, pp. 193-205, Jul. 2019.
[13] J. Yim and K. A. Sohn, "Enhancing the performance of convolutional neural networks on quality degraded datasets," in Int. Conf. Digit. Image Comput. Tech. Appl., DICTA'17, 8 pp., Sydney, NSW, Australia, 29 Nov.-1 Dec. 2017.
[14] I. F. Jafar, R. A. Alna'Mneh, and K. A. Darabkh, "Efficient improvements on the BDND filtering algorithm for the removal of high-density impulse noise," IEEE Trans. Image Process., vol. 22, no. 3, pp. 1223-1232, Nov. 2013.
[15] S. Q. Yuan and Y. H. Tan, "Impulse noise removal by a global-local noise detector and adaptive median filter," Signal Processing, vol. 86, no. 8, pp. 2123-2128, Aug. 2006.
[16] D. Ze-Feng, Y. Zhou-Ping, and X. You-Lun, "High probability impulse noise-removing algorithm based on mathematical morphology," IEEE Signal Process. Lett., vol. 14, no. 1, pp. 31-34, Jan. 2007.
[17] K. S. Srinivasan and D. Ebenezer, "A new fast and efficient decision-based algorithm for removal of high-density impulse noises," IEEE Signal Process. Lett., vol. 14, no. 3, pp. 189-192, Mar. 2007.
[18] S. Esakkirajan, T. Veerakumar, A. N. Subramanyam, and C. H. PremChand, "Removal of high density salt and pepper noise through modified decision based unsymmetric trimmed median filter," IEEE Signal Process. Lett., vol. 18, no. 5, pp. 287-290, Mar. 2011.
[19] G. Yu, S. Niu, and J. Ma, "A hybrid spectral gradient method for removing salt-and-pepper impulse noise," in Proc. 4th Int. Congress on Image and Signal Processing, pp. 765-768, Shanghai, China, 15-17 Oct. 2011.
[20] P. Shanmugavadivu and P. S. Eliahim Jeevaraj, "Fixed-value impulse noise suppression for images using PDE based adaptive two-stage median filter," in Proc. Int. Conf. on Computer, Communication and Electrical Technology, ICCCET'11, pp. 290-295, Tamilnadu, India, 18-19 Mar. 2011.
[21] X. Zhang and Y. Xiong, "Impulse noise removal using directional difference based noise detector and adaptive weighted mean filter," IEEE Signal Process. Lett., vol. 16, no. 4, pp. 295-298, Apr. 2009.
[22] S. Sadeghi, A. Rezvanian, and E. Kamrani, "An efficient method for impulse noise reduction from images using fuzzy cellular automata," AEU-Int. J. Electron. Commun., vol. 66, no. 9, pp. 772-779, Sept. 2012.
[23] D. Androutsos, K. N. Plataniotis, and A. N. Venetsanopoulos, "Novel vector-based approach to color image retrieval using a vector angular-based distance measure," Comput. Vis. Image Underst., vol. 75, no. 1, pp. 46-58, Jul. 1999.
[24] T. Toffoli and N. Margolus, Cellular Automata Machines: A New Environment for Modeling, 1987.
[25] S. S. Wang and C. H. Wu, "A new impulse detection and filtering method for removal of wide range impulse noises," Pattern Recognit., vol. 42, no. 9, pp. 2194-2202, Sep. 2009.
[26] P. Y. Chen and C. Y. Lien, "An efficient edge-preserving algorithm for removal of salt-and-pepper noise," IEEE Signal Process. Lett., vol. 15, pp. 833-836, Dec. 2008.
[27] C. Szegedy, et al., "Going deeper with convolutions," in Proc. of the IEEE Computer Society Conf. on Computer Vision and Pattern Recognition, CVPR'15, 9 pp., Boston, MA, USA, 7-12 Jun. 2015.
[28] J. Gu, et al., "Recent advances in convolutional neural networks," Pattern Recognit., vol. 77, pp. 354-377, May 2018.
[29] A. K. Samantaray, P. Kanungo, and B. Mohanty, "Neighbourhood decision based impulse noise filter," IET Image Process., vol. 12, no. 7, pp. 1222-1227, Jul. 2018.
[30] K. Simonyan and A. Zisserman, "Very Deep Convolutional Networks for Large-Scale Image Recognition," Sep. 2014.
[31] O. Russakovsky, et al., "ImageNet large scale visual recognition challenge," Int. J. Comput. Vis., vol. 115, no. 3, pp. 211-252, Dec. 2015.
[32] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," Proceedins of the IEEE, vol. 86, no. 11, pp. 2278-2324, Nov. 1998
[33] Y. Wang, A. Szlam, and G. Lerman, "Robust locally linear analysis with applications to image denoising and blind inpainting," SIAM J. Imaging Sci., vol. 6, no. 1, pp. 526-562, Jan. 2013.
[34] B. Dong, H. Ji, J. Li, Z. Shen, and Y. Xu, "Wavelet frame based blind image inpainting," Appl. Comput. Harmon. Anal., vol. 32, no. 2, pp. 268-279, May 2012.
[35] M. Yan, "Restoration of images corrupted by impulse noise and mixed gaussian impulse noise using blind inpainting," SIAM J. Imaging Sci., vol. 6, no. 3, pp. 1227-1245, Jan. 2013.
[36] J. Salmon, Z. Harmany, C. A. Deledalle, and R. Willett, "Poisson noise reduction with non-local PCA," J. Math. Imaging Vis., vol. 48, no. 2, pp. 279-294, Feb. 2014.
