双树复小波变换域矿区遥感图像自适应滤波

Adaptive Filtering of the Remote Sensing Image of Mining Area in Dual-tree Complex Wavelet Transform Domain

矿区遥感图像因受成像环境、成像器件固有缺陷等因素的影响容易出现不同程度的失真,为此,结合双树复小波变换(Dual-tree complex wavelet transform,DTCWT)多尺度图像分析的优良特性,提出了一种矿区遥感图像自适应滤波算法。首先对获取的视觉效果不佳的遥感图像进行直方图均衡化处理,使得增强后的图像灰度分布较为合理,提高图像的对比度;然后对增强后的图像进行双树复小波变换,对获得的高频分解系数采用改进的多级中值滤波算法进行处理;最后,将低频分解系数与滤波后的高频分解系数进行逆双树复小波变换。其中改进的多级中值滤波算法相对于经典多级中值滤波算法进行了2点改进:1将原有的4个方向滤波窗口扩展为7个,更有利于保持图像中信息的多方向特性;2对新增设的3个滤波窗口分别进行加权中值滤波,将上述7个滤波窗口的滤波值采用一种基于图像灰度值相关性的判别方法进行处理,剔除与待滤波像素点相关性不强的滤波值,将剩余的滤波值计算均值输出;MATLAB平台试验结果表明:新算法的总体性能相对于经典多级中值滤波、中值滤波、双边滤波等算法而言,优势较为明显。

Remote sensing image is easily influenced by the factors such as imaging environment,the inherent defects of imaging device during the process of imagining,the phenomenon of distortion with different degrees of the remote sensing image is appeared. Combing with the excellent characteristics of multi-scale image analysis of dual-tree complex wavelet transform（ DTC-WT）,a adaptive filtering algorithm of remote sensing image is proposed. Firsly,the obtained remote sensing image with poor visu-al effect is processed by histogram equalization algorithm to improve the image contrast;then,the enhanced remote sensing image is conduct dual-tree complex wavelet transform,the low-frequency coefficients and high-frequency coefficients are obtained,the low-frequency coefficient is remained unchanged,the high-frequency coefficients are filtered by the improved multi-stage median filtering algorithm,the classical multi-stage median filtering algorithm is improved as follows：①the four direction filtering win-dows of the classical multi-stage median filtering algorithm are extended to seven direction filtering windows,which is more ad-vantages to keep the multiple directions characteristics of the information in remote sensing image;②the new added three filtering windows are denoised by weighted median filtering algorithm,the filtering values of the above seven filtering windows are pro-cessed by a discriminant method based on image gray value relevance so as to eliminate the filtering values with poor correlation to the filtering values of the pixels points,the average value of the rest of the filtering values is regarded as the filtered value;fi-nally,the low-frequency coefficient and the filtered high-frequency coefficients are conducted inverse dual-tree complex wavelet transform. The experimental results based on MATLAB software show that：the algorithm proposed in this paper maintain the in-tegrity of the detail information of the remote sensing image of mining area during the process of filtering,the performance the al-gorithm has a certain degree of ascension related to the classical multi-stage median filtering algorithm,besides that,the perform-ance of the algorithm has obvious advantages to the algorithms of median filtering and bilateral filtering.