递进的迭代脉冲噪声检测

目的 脉冲噪声是引起图像质量下降的主要原因,其滤除工作一直是图像处理领域的研究热点。对现行开关滤波算法在脉冲噪声检测时间、检测精准度和恢复策略上存在的问题进行理论分析,提出一种递进的迭代脉冲噪声检测算法(PIND),使噪声图像能够获得更好的恢复效果。方法 首先,采用具有全局统计意义的灰度直方图确定脉冲噪声与真实像素之间的灰度值的上边界和下边界,根据这个界线区分出疑似点和真实点;然后,利用具有局部结构意义的方法将噪声点从疑似点中寻找出来并判断噪声类型,存储在决策表G中;最后,根据决策表G中存储的噪声类型信息采用3种不同的的恢复策略滤除噪声。结果 对Lena、Peppers和Monkey 3幅具有代表性的图像增加不同密度和尺度的噪声进行对比实验,得出的数据表明,本文算法的脉冲噪声检测时间比现行两种经典算法提高520倍和15倍;检测精准度比现行经典开关滤波算法更加精准,准确率可以达到99%以上;恢复图像也具有更好的视觉效果和12 dB的峰值信噪比(PSNR)提升。结论 提出递进的迭代脉冲噪声检测算法能够在有效滤除脉冲噪声的同时,充分保护图像细节和恢复图像原有特征,并能够在噪声检测时间和精度以及峰值信噪比上弥补现行开关滤波算法的不足。

Progressive iterative impulse noise detection

Objective Impulse noise is the main cause low image quality. The filtering of impulse noise has always been a research hotspot in the field of image processing. On the basis of theoretical analysis for current switching filtering algorithms in terms of detection time, detection accuracy, and recovery strategy, this study proposes a progressive iterative impulse noise detection algorithm, which can obtain a high recovery effect from noisy images.Method First, we adopt gray-level histograms that possess global statistical significance to identify the pixel gray value boundary of impulse noise and real pixels b₁, b₂. By using these histograms, we can distinguish the suspected points and real points. Second, we use the method of local structure significance to identify and classify the noise points from the suspected points. These points are then saved in Table G. Finally, according to the different noise types in Table G,we use 3 different strategies to remove the noise points. Result The experiments on three representative images with different noise densities and noise intensities show that our detection time is 520 times and 15 times faster than that of the two current classic algorithms, respectively. Furthermore, the proposed method has a detection accuracy of 99%, recover images with excellent visual effects, and enhances the peak signal-noise ratio to 12 dB.Conclusion The proposed algorithm can protect the image detail and recover the original features of the image when filtering impulse noise. The proposed method can also compensate for the disadvantage of current switching filters in terms of detection time, detection accuracy, and peak signal-noise ratio.