基于边窗的共现变分保边平滑算法

Co-occurrence Variational Edge-preserving Image Smoothing based on Side Window

摘要:

针对图像平滑过程中边缘模糊及小尺度结构丢失的问题，提出了一种基于边窗的共现变分保边平滑算法；该算法通过共现滤波方法，利用其核心思想共现矩阵的特性，能够为结构区域与纹理区域的梯度差异提供更准确的衡量，进一步提高相对权重的差异；提出一种改进的边窗高斯滤波器，结合方向一致性原理，设计了新的边窗选择机制，能够更全面地选择合适的边窗；首次将改进的边窗高斯滤波与全局优化框架结合，在结构区域的边缘处表现出更强的引导，更倾向于抑制纹理；经实验验证，所提算法在平滑效果、边缘保留以及小尺度结构保持方面优于经典算法与先进算法，具备较强鲁棒性和优越性；此外，所提算法在SSIM和GMSD指标上均取得最优值0.99445和0.0010324，在PSNR和SCOOT指标上取得次优值40.2448和0.77645。

Abstract:

Aiming at the issues of edge blurring and small-scale structure loss during image smoothing, a co-occurrence variational edge-preserving image smoothing based on side window is proposed; the algorithm adopts a co-occurrence filtering method and the characteristics of the co-occurrence matrix, which is its core idea, to provide a more accurate measure of the gradient difference between the structural region and the textural region, and further improves the difference of the relative weights; an improved side window Gaussian filter is proposed. Combined with the principle of directional consistency, a novel side window selection mechanism is designed, which is able to select suitable side windows more comprehensively; the improved side window Gaussian filter is combined with the global optimization framework for the first time, which exhibits stronger guidance at the edges of the structural region and is more inclined to suppress the textures; experimental verification shows that the proposed algorithm is superior to classical algorithms and advanced algorithms in terms of smoothing effect, edge preservation and small-scale structure preservation, and has strong robustness and superiority; in addition, the proposed algorithm achieves the optimal values of 0.99445 and 0.0010324 in SSIM and GMSD metrics, and the suboptimal values of 40.2448 and 0.77645 in PSNR and SCOOT metrics.