基于背景最佳滤波尺度的红外图像复杂度评价准则

提出一种基于背景最佳滤波尺度的红外图像复杂度评价准则来解决传统方法评价背景效果较差的问题. 同时, 这种方法还可以为红外图像滤波提供最佳高通滤波尺度信息, 从而对红外图像进行性能最佳滤波. 首先, 生成高斯仿真目标并与红外图像进行融合, 获得包含仿真目标及真实红外背景的图像. 然后, 在不同高斯滤波尺度下对图像滤波, 并计算滤波后仿真目标的信噪比. 最后, 取滤波后目标信噪比最大时的滤波尺度作为背景最佳滤波尺度, 使用该尺度可评价红外图像的复杂度. 另外, 本文还使用数学模型推导了红外图像最佳滤波尺度, 得出最佳滤波尺度的数学表达式. 大量实验表明: 1) 本文推导的最佳滤波尺度数学表达式与实验曲线吻合. 2) 这种方法在评价红外图像复杂度方面比传统的基于信息熵的方法效果要好很多. 并且这种方法获取的红外背景复杂度为滤波最佳尺度, 可以直接利用这项指标对图像进行最佳滤波从而更好地检测弱小目标. 3) 仿真目标尺度越大, 最佳滤波尺度也会相应增大. 因此, 在评价图像复杂度时, 应使用相同尺度的仿真目标, 不同图像之间才具备可比性. 同时, 最佳滤波尺度与仿真目标的强度无关. 4) 本文算法使用的滤波器宜用高斯及Butterworth高通滤波器实现. 5) 本文提出的方法不仅可以有效分析红外视频的复杂度, 并且可以通过复杂度的变化分析图像内容的突变.

An evaluation criterion of infrared image complexity based on background optimal filter scale

An evaluation of infrared image complexity is proposed based on the background optimal filtering to solve the problem that the traditional methods have given poor results in the background evaluation. Meanwhile, the optimal filtering scale for infrared image filtering can be given by this method, it will provide a guidance for optimal infrared image filtering. First, we generate the Gaussian simulated target and fuse it to the infrared image to obtain the real infrared image with the simulated target. Then, this image is filtered in different scales and the signal-to-noise ratio of the target after filtering is calculated. Finally, the maximal value of signal-to-noise ratio of the target is used as the background optimal filter scale, to evaluate the infrared image complexity. Besides, the infrared filtering scale is deduced by establishing the mathematic model, and then the mathematical expression of optimal filtering scale is obtained. A lot of experiments indicate that: 1) The mathematical expression of optimal filtering scale agrees with the experimental results. 2) The result of our method is better than that of the traditional methods based on information entropy. Because the optimal filtering scale is obtained by using our method, we can use this scale to filter the infrared image to effectively detect a small target. 3) When the scale of simulated target increases, the optimal filtering scale increases accordingly. So, when we calculate the infrared image complexity, the scale of simulated target must be the same. We can compare the infrared image complexity between different images. Moreover, the optimal filtering scale is independent of the intensity of simulated target. 4) The effect of Gaussian and Butterworth high-pass filter is better than that of the ideal high-pass filter in the proposed method. 5) The infrared image complexity can be analyzed by the proposed method effectively. Moreover, changes of different image contents can be analyzed by using the optimal filtering scale.