基于改进的直方图均衡化与边缘保持平滑滤波的红外图像增强算法

INFRARED IMAGE ENHANCEMENT ALGORITHM BASED ON IMPROVED HISTOGRAM EQUALIZATION AND EDGE PRESERVING SMOOTHING FILTERING

为了提高红外图像的全局与局部对比度,并有效抑制背景噪声,提出基于改进的直方图均衡化与边缘保持平滑滤波的红外图像增强算法。引入边缘保持平滑滤波,将低质量红外图像分解为一个低频分量和一个高频分量序列;基于模糊统计理论,利用红外图像的强度等级的不确定性,形成平滑直方图,确定出局部最大值,利用优化的平台直方图对低频分量完成增强;根据高频分量的标准差,将其高频分量分类为强边缘、中边缘与弱边缘系数,再设计三个不同的增强方法,对这三类系数进行差异增强,从而得到增强的高频分量序列;将增强的低频分量与一系列的高频分量完成组合,形成增强图像;引入非局部均值滤波,对增强图像实施降噪处理。测试结果表明:与当前低质量红外图像增强方案相比,该方法拥有更高的增强视觉质量,更好地兼顾全局与局部对比度,消除过渡增强、伪影与噪声,且输出图像具有更大的熵值与标准偏差值,分别保持在6.8、5.3以上。

In order to improve the global and local contrast of infrared image and effectively suppress the background noise, we proposed an infrared image enhancement algorithm based on improved histogram equalization and edge preserving smoothing filtering. We introduced an edge preserving smoothing filtering to decompose a low-quality infrared image into a low-frequency component and a high-frequency component sequence. Then, based on the fuzzy statistical theory, a smoothing histogram was formed by using the uncertainty of the intensity level of the infrared image to determine the local maximum, and the optimized platform histogram was used to enhance the low frequency components. According to the standard deviation of high-frequency components, the high-frequency components were classified as strong edge, middle edge and weak edge coefficient, and three different enhancement methods were designed to differently enhance the three coefficients for obtaining the high frequency component sequence. The enhanced low frequency component was combined with a series of high frequency components to form an enhanced image. We also introduced nonlocal mean filtering to reduce noise in the enhancement image. Test results show that the proposed method has higher visual quality, better global and local contrast than the current low-quality infrared image enhancement scheme. This method can eliminate transition enhancement, artifacts and noise, and the output images have greater entropy and standard deviation values which maintain at 6.8 and 5.3.