为了改善微光情况下可见光图像传感器输出图像的质量,提出了一种基于高斯混合模型的自适应微光图像增强算法.对图像的直方图进行混合高斯建模,应用改进的期望最大化算法对直方图拟合,从而获取高斯混合模型的最优参数,然后根据各个聚类的交点将直方图分区,最后确定输出图像所属聚类的映射关系,同时应用保持最大熵方法逼近人类视觉特性映射函数得到最终的增强图像.实验结果表明,此图像增强模型能自适应确定最佳聚类个数,提高直方图拟合的运算速度,一帧图像平均处理时间为0.37 s,在相关信息熵和纹理信息等的客观评价中,增强结果明显优于传统方法,有效地提高了微光图像的对比度,同时保持了图像的细节.

In order to improve the output quality of visible light sensor in low-light environment,an adaptive image enhancement algorithm using Gaussian mixture modeling is proposed in this paper.The histogram of image is modeled with Gaussian mixture modeling and the improved EM algorithm is used to fit the histogram and get the best parameters.Then,the histogram is separated into sub-histograms based on the intersections of Gaussian components.Finally,the mapping is achieved according to the parameters of output image,and the final enhanced image is obtained by the maximum entropy preserving method which tends to the characteristics of human visual.The experimental results show that the algorithm can determine the optimal number of clusters adaptively and improve the speed of the histogram fitting which costs 0.37 s averagely.Comparing with traditional methods,the enhancement result is superior in terms of objective evaluations of related information entropy and texture information.It can improve the contrast of the low light level image and maintain the details.