Images captured during the manufacturing of iron green pellets may be seriously degraded by haze, which greatly affects the performance of machine vision systems for pellet size measurement. Hence, to improve the quality of green pellet images and pellet size measuring accuracy, a fast image dehazing method is proposed in this study. By analyzing the different features between haze and haze-free image patches, a new and simple prior, which is that haze-free image patches have the maximal local image entropy, is proposed. The transmission map is then calculated using the proposed prior and the haze-free image is obtained. The dehazing effect of the proposed algorithm was assessed using qualitative and quantitative criteria, and compared with six current image dehazing algorithms on two datasets of hazy green pellet images established by ourselves. The results demonstrate that the proposed algorithm achieves better dehazing performance than the other algorithms and can improve the accuracy of pellet size measurement. Moreover, the computing efficiency of the proposed method meets the requirements for real-time industrial applications when used as a pre-processing step. This proposed dehazing method will enable machine vision monitoring systems to operate in industrial environments where dust and steam are common.

在制造铁绿色颗粒时捕获的图像可能会因雾霾而严重劣化，这会极大地影响用于颗粒尺寸测量的机器视觉系统的性能。因此，为提高绿色颗粒图像的质量和颗粒尺寸的测量精度，本研究提出了一种快速的图像去雾方法。通过分析雾度和无雾度图像块之间的不同特征，提出了一种新的简单方法，即无雾度图像块具有最大的局部图像熵。然后使用建议的先验值计算透射率图，并获得无雾图像。使用定性和定量标准评估了该算法的除雾效果，并与我们自己建立的两个模糊绿色颗粒图像数据集上的六种当前图像除雾算法进行了比较。结果表明，与其他算法相比，该算法具有更好的除雾性能，可以提高颗粒尺寸的测量精度。此外，所提方法的计算效率在用作预处理步骤时可以满足实时工业应用的要求。这种提议的除雾方法将使机器视觉监控系统能够在粉尘和蒸汽普遍的工业环境中运行。