材料期刊网

目的有效去除生产现场所采集的带钢图像上的混合噪声。方法结合Shearlet变换具有较好的稀疏表示图像特征的性质与全变分各向异性扩散的优点,提出一种带钢图像去噪新算法。对Shearlet变换分解后的图像进行硬阈值处理,再进行Shearlet变换重构形成估计图像,采用改进自适应的变差正则化的极小化迭代模型对估计图像进行迭代修正。结果去噪后的图像具有很好的视觉效果,避免了伪吉布斯效应的产生。在强噪水平下,对比新模型与小波去噪,PSNR提高了约9 dB,均方差降低了约319。结论该方法获得了较好的峰值信噪比增益,使信号幅度有较高的保真度,具有更好的平滑噪声和边缘保持功能。

ABSTRACT:Objective To effectively remove mixed noise from the image of acquisition steel strip in the production field. Methods Combining the advantage of the Shearlet transform which has better properties to sparsely express the characteristics of the images and the total variational anisotropic diffusion, a new image denoising model was proposed. After Shearle transform decompo-sition, the image was processed by hard thresholding, and then the estimated image was formed after Shearle transform reconstruc-tion. The algorithm used iterative model of minimization of total variation regularization to correct the estimated image. Results The denoised image had good visual effect, and the creation of pseudo Gibbs effect was avoided. The comparison of the new model with wavelet denoising under the strong noise level showed that PSNR was increased by 9 dB and MSE was reduced by 319. Conclusion Numerical examples demonstrated that this method could achieve better PSNR gain, and the results showed that the filters had high fidelity of signal amplitude, and better function in smoothing noise and preserving edges.