Ying Cui,∗ Shixin Li,∗ Decai Qu,∗ Xiaoyu Fan,∗ and Hongchao Lu∗∗
Hopfield neural network, median filter, wavelet denoising method,image processingNomenclatureE network energyf(m, n) binary function of imageg decomposition scale∗ College of Electromechanical Engineering, Qingdao Univer-sity of Science and Technology, Qingdao, 266061, China;e-mail: [email protected], {seawolf lsx, qudcmail, fxy1321390481}@163.com∗∗ School of Civil Engineering & Transportation, South ChinaUniversity of Technology, Guangzhou, 510641, China; e-mail:luhclhc@
Hopfield neural network, a single-layer recurrent neural network, has been effectively applied to data recognition for handwritten characters, letters, and digits due to its considerable fault toler- ance and self-adaptability in recent times. However, it would fail to achieve satisfactory outputs once the images are contaminated with serious noise, which severely affects its application in image processing. In order to overcome this defect, this paper proposes an improved Hopfield Neural Network which utilizes the combination of median filtering and two-dimensional wavelet transform as a before- hand denoising method. Simulations are conducted to estimate the effectiveness of the improved algorithm. Subjective and objective results, such as visual images and peak signal-to-noise rate values, have been presented for the purpose of evaluating the denoising performance and the memory property under noise conditions. The results show that the combinational denoising strategy shows supe- rior performance in noise removal and feature details preservation, contributing to improving the memory capacity of Hopfield network significantly. Meanwhile, the improved algorithm shows reliable and consistent performance for different digital images at low-to-high noise levels.
Important Links:
Go Back
