针对红外图像中常出现的颗粒噪声，在对自适应中值滤波（AMF）改进的基础上，提出了一种基于噪声检测的加权融合中值滤波算法。该算法对于噪声的滤除分为检测和滤波2个环节。首先根据对大量图像信息分布特征进行分析，设计出了较为通用的4类噪声检测模版；其次分别采用各模版对噪声进行极值检测并加以标记，以区分出极值点与像素点；然后根据模版内极值点与非极值点的数目比例确定噪声点，并将非极值点取均值作为噪声滤波结果输出；对于区域的极值点，引入距离判别准则加以确认，被确认为噪声点的像素值通过模版内非极值点分别取均值和中值并加权融合得到；最后对4个模版处理后的滤波值进行等权融合，得到最终的滤波图像。分别与经典中值滤波（MF）、自适应中值滤波（AMF）、开关中值滤波（SMF）性能比较，实验表明，本文滤波算法对于高强度的斑点噪声具有较好的滤波性能，对于红外图像处理具有一定的借鉴价值。%A new improved median filtering algorithm is presented so as to remove the salt&pepper noise in infrared image effectively on the basis of the adaptive median filter(AMF). The filtering algorithm is divided into noise detection and noise filter. Firstly, according to the analysis of the characteristics of the distribution of the digital image information, four types of noise detection template are designed. Then, the new templates are used respectively to detect the extreme values and label them in the image, so as to distinguish between the extreme value pixels and non-extreme value pixels. Thirdly, the noise pixels are determined according to the ratio between extreme value pixels and non-extreme value pixels. The average value of the non-extreme value pixels is calculated and regarded as the filtering results. The distance criterion is introduced to deal with the extreme value point in large area in the template, and the value of the noise pixels is replaced by the weighted fusion value of the mean value and average value of the non-extreme pixels value in the template. Finally, the filtering image is obtained by weighted fusion for the results of the four types of templates. Experimental results indicate that the proposed filtering algorithm outperforms many other algorithms such as median filter, adaptive median filter and switch median filter, which allows for its application in real-time applications.
展开▼
