中文摘要：

由于煤矿井下空间环境受粉尘等因素的影响,视频监控系统获取的作业环境图像存在照度低、照度不均匀等问题,而且现有的算法在处理煤矿非均匀照度图像时会出现颜色失真或者过增强现象,这不利于对图像的判读和应用。结合煤矿的实际数据,提出了一种新的图像增强算法,克服了现有算法存在的问题：分析井下视频图像特点,采用具有边缘保持特性的多尺度引导滤波获取照度分量;基于Retinex理论,将图像分解为照度分量与反射分量;针对照度不均匀的特性,提出一种新的＂S型＂曲线函数对其进行调整;分析图像的特性,引入受限对比度自适应直方图对其对比度进行增强;提出新的图像增强模型,利用细节增强系数和照度增强系数实现对图像的综合增强。与其他4种算法的对比试验表明,本文算法在主观评价和客观评价方面都优于上述算法。实验证明,本算法具有有效提升图像整体亮度和对比度,同时避免光源附近亮区域的过增强现象的特点,能够满足矿山实际应用需求。

英文摘要：

Due to the environment of coal mine space is affected by grime, the images obtained by video surveillance have low illumination, non-uniform illuminance problems, and the existing algorithms cause color distortion or excessive enhancement in dealing with non-uniform illumination mine image, which is not conducive to the interpretation and ap- plication of images. With the actual images data, this paper proposed a new non-uniform image enhancement algorithm based on illumination adjustment combining with Guided Filtering and ＂S Curve＂ function, which may overcome the problems of the existing algorithms. By analyzing the characteristics of the video images, a multi-scale guide filtering, which can preserve the edges and corners, is adopted to get the illumination component. Then based on Retinex theory, the image is decomposed into an illuminance component and a reflection component. Regarding the non-uniform illumi- nation, a new S-shape curve function is proposed to adjust the illuminance component. Considering the image has a rel- ative low contrast, the contrast limited adaptive histogram is further used to enhance the contrast. Finally, a new image enhancement method is proposed with the detail enhancement coefficient and the illumination enhancement coefficient achieve a better enhancement performance of non-uniform images. Compared with the other four algorithms, this papershows that the proposed algorithm is superior to the above algorithms both in subjective evaluation and objective evalu- ation. Finally, this algorithm can effectively enhance the overall image brightness and contrast, while avoiding excessive enhancement phenomenon of the bright region generated near the light, showing the superiority of the algorithm.