(1. 中南大学 信息科学与工程学院，长沙 410083；
2. 湖南文理学院 电气与信息工程学院，常德 415000)
摘 要: 针对矿物浮选过程中泡沫图像易受噪声影响，存在纹理细节模糊、灰度值对比度低等问题，提出一种浮选泡沫图像的非线性降噪方法。首先构造一种改进方向波变换，保证信号的平移不变性，同时采用提升算法减小其运算量。然后通过对分解系数建模，针对低频子带系数采用多尺度Retinex算法进行处理，以改善整体亮度均匀性，提高对比度；对各高通子带构建基于高斯混合尺度模型的分解系数邻域模型，并利用Bayes最小均方(BLS)估计进行局部去噪。最后利用所提出的方法对大量浮选泡沫图像进行去噪分析。结果表明：所提出的降噪方法能突出泡沫图像的纹理细节信息，提高泡沫图像的对比度，在信噪比和实时性上有明显提高，为后续泡沫图像的分割和工况识别奠定基础。
关键字: 浮选泡沫图像；图像降噪；方向波变换；高斯混合尺度模型；Retinex算法
（1. School of Information Science and Engineering, Central South University, Changsha 410083, China;
2. College of Electrical and Information Engineering, Hunan University of Arts and Science, Changde 415000, China）
Abstract:Considering the defects, such as easy sensitivity to noise and heavy texture, low contrast of gray value in the process of the floatation of foam image, a non-linear de-noising method was proposed. Lifting improved directionlet transform was firstly constructed, which not only ensured the shifting invariance but reduced its complexity. Multi-scale Retinex algorithm dealing with low-frequency subband coefficient was proposed for improving luminance uniformity and overall contrast. For high-pass subband, a model of decomposition coefficients neighbourhood based on Gaussian scale mixtures model was proposed for de-noising the image locally using Bayes least square (BLS). The analysis on the effect of de-noising was given to lots of real froth images. The results show that the proposed method is successful in maintaining edges and is superior in de-noising in term of PSNR and visual effect. It lays a foundation for foamy segmentation and analyzing grade from flotation froth image.
Key words: flotation froth image; image de-noising; directionlet transform (DT); Gaussian scale mixture (GSM); Retinex algorithm
