基于模糊核和梯度分布的部分模糊图像检测

Research and improvement of blur kernel evaluation and gradient distribution in the partial blur image detection

针对传统模糊核评价方法不能区分模糊平滑区域和焦点平滑区域,模糊图不易于观察的问题,进行了研究。首先对图像块进行模糊核评价,得到其初始模糊度和模糊图。然后计算图像块梯度分布相似性系数,由梯度分布相似性系数来优化模糊图。在实验数据集上的测试表明,改进后的方法适用于失焦模糊图像和运动模糊图像,能更好的捕获部分模糊图像中的清晰区域,与许多方法相比,算法耗时少,清晰区域检测更加准确。

The traditional method depends on blur kernel evaluation method which leads to any problems,it is difficult to distinguish the blur smooth region and the in-focus smooth region,inconvenient to observe the blur map,so this paper researches on it. Firstly,each image block blur kernel was evaluated and it got initial blur degree and blur map. Then, the gradient distribution similarity coefficients of image blocks were calculated,the blur maps were optimized by the the gradient distribution similarity coefficients. The test on the experimental data set shows that the improved method is suitable for the blurred image and motion blurred image,which can capture the clear region of the blurred image and the motion blurred image better. The algorithm takes less time,the detection of clear areas is more accurate compared to the various methods.