图像去噪过程可视化系统设计与实现

Design and Implementation of a Visualization System for Image Denoising Process

摘要: 传统图像去噪算法通常以最终输出结果作为主要评价依据，其内部迭代优化与噪声抑制过程对用户呈“黑箱”状态，难以支撑算法原理教学与效果分析。针对这一问题，本文设计并实现一套面向教学与演示的图像去噪过程可视化系统，集成均值滤波、中值滤波、高斯滤波、BM3D等多种经典算法，通过迭代中间帧捕获与动态GIF合成，将抽象的去噪过程转化为直观可视的动画演示。在此基础上，本文提出一种参数自适应的交替迭代去噪算法，对纯椒盐噪声与椒盐噪声和高斯噪声的混合噪声在每轮先进行中值滤波，再用高斯滤波与双边滤波交替执行；对纯高斯噪声直接进行高斯–双边交替去噪。滤波参数由噪声估计、梯度复杂度因子与迭代退火因子联合调节，在有效去除噪声的同时更好地保留边缘与细节信息。系统采用多线程架构分离计算任务与界面渲染，保证交互流畅，并通过PSNR、SSIM实现去噪质量的客观量化评价。实验结果表明，该系统能够清晰展示各类算法的去噪演化过程，本文所提出参数自适应的交替迭代去噪算法在多种噪声类型与噪声强度下均具有更优综合表现，可为图像去噪算法的教学演示与效果对比提供有效支撑。

Abstract: Traditional image denoising algorithms typically take the final output result as the primary evaluation criterion, leaving the internal iterative optimization and noise suppression process as a “black box” to users, which makes it difficult to support the teaching of algorithm principles and performance analysis. To address this issue, this paper designs and implements a visualization system for the image denoising process oriented toward teaching and demonstration. The system integrates multiple classical algorithms, including Mean Filtering, Median Filtering, Gaussian Filtering, and BM3D. By capturing intermediate iteration frames and synthesizing dynamic GIFs, the abstract denoising process is transformed into an intuitive visual animated demonstration. On this basis, this paper proposes a parameter-adaptive alternating iterative denoising algorithm. For pure salt-and-pepper noise and mixed noise consisting of salt-and-pepper and Gaussian noise, median filtering is first performed in each iteration round, followed by alternating execution of Gaussian filtering and bilateral filtering; for pure Gaussian noise, Gaussian-bilateral alternating denoising is directly performed. The filtering parameters are jointly regulated by noise estimation, a gradient complexity factor, and an iterative annealing factor, which effectively removes noise while better preserving edge and detail information. The system adopts a multi-threaded architecture to separate computational tasks from interface rendering, ensuring smooth interaction, and provides objective quantitative evaluation of denoising quality through PSNR and SSIM metrics. Experimental results demonstrate that the system can clearly present the denoising evolution processes of various algorithms, and the proposed parameter-adaptive alternating iterative denoising algorithm achieves superior comprehensive performance under multiple noise types and intensities, which can provide effective support for the teaching demonstration and performance comparison of image denoising algorithms.