基于时空曲率正则化的细胞图像序列去噪

目的显微细胞成像系统获取的图像序列由于光照、电磁干扰等因素的影响,不可避免地存在一定程度的噪声,消除噪声得到清晰的细胞图像是后续细胞形态特征提取和分析的首要步骤。本文引入细胞图像序列的时域信息来构建时空曲率正则化约束,以实现细胞图像序列的去噪处理。方法首先,利用细胞图像序列的空域和时域相关性,构建基于时空曲率正则化的图像序列去噪模型;然后,通过增广拉格朗日乘子法实现模型的优化求解;最后,分别通过对叠加有不同高斯白噪声水平的纤维母细胞和多能干细胞图像序列进行去噪实验,以验证去噪效果。结果与总变分去噪法、三维阈值剪切去噪法和空间曲率正则化去噪法相比较,基于时空曲率正则化的细胞图像序列去噪方法应用于2组细胞图像序列去噪的视觉效果,及峰值信噪比(peak signal to noise ratio,PSNR)和结构相似度(structural similarity,SSIM)都优于其他3种方法。结论与其他3种去噪方法相比,此方法更加充分利用了细胞图像序列的时域信息,去噪后能有效地维持图像对比度,振铃效应不明显,对高斯噪声具有更好的适应性和稳定性,可应用于细胞形态变化检测的前期处理阶段。

Cell image sequence denoising based on spatio-temporal curvature regularization

Objective Affected by some factors,such as light,electromagnetic interference and so on, there is a certain degree of noise in the cell image sequence obtained by microscopic cell imaging system, inevitably. To get clear cell images by denoising,it is the most important for the subsequent cell morphological feature extraction and analysis steps. To eliminate the noise of cell image sequence, we introduce the time domain information of cell image sequence to build a Spatio-temporal curvature regularization constraint. Methods First,introducing the correlation of space domain and time domain in cell image sequence,we built an image sequence denoising model based on spatio-temporal curvature regularization. Then, the optimization solution of this model was solved through the augmented Lagrangian method. Finally, in order to verify the denoising effect, the denoising experiments were performed on Fibroblast and pluripotent stem cell image sequence adding with white Gaussian noise. Results Compared with total variational denoising,3D threshold shear denoising and spatial curvature regularization denoising,this method was superior to other three methods on the visual effect of denoising,peak signal to noise ratio ( PSNR ) and structural similarity ( SSIM ) . Conclusions Compared with three other denoising method based Spatio-temporal volume, this method makes full use of the temporal information of cell image sequence,and can maintain the image contrast and has a better adaptability and stability with respect to Gaussian noise, and then the ringing effect is not obvious. This method can be applied to the cell morphological change detection in early processing stage.