面向工业检测的图像快速去直线运动模糊方法

Fast moving line motion de-blurring for image detection of industrial inspection

摘要:

在基于机器视觉的工业检测中，由于受到传送带速度变化或曝光不足的影响，得到的图像有时会产生运动模糊，影响检测效果.为去除工业检测中的图像运动模糊，提出了基于R-L引导滤波的快速去直线运动模糊方法.首先对图像的模糊参数进行评估，使用图像的频谱图积分评估图像的模糊方向，使用微分自相关方法得到模糊图像尺寸；然后通过提出的R-L引导滤波方法快速去除运动模糊并抑制振铃效应，得到清晰图像.通过对工件图像去运动模糊的仿真，证明了模糊核估计的准确性；与TPKE和HQ比较，提出的去模糊方法在对噪声的鲁棒性和抑制振铃的效果更好.通过对不同速度、不同工件下实拍图像的去运动模糊实验，以及客观评价指标、去模糊前后尺寸测量的差异和运行速度证明了该方法在工业检测中的优势.实验结果表明：提出方法的R_PSNR和R_SIMM优于对比方法，在抑制振铃的同时保持了边缘的清晰; 同时，提出方法对尺寸的测量影响也更小，在小尺寸的工件上具有优势，最大误差为0.31像素.

Abstract:

In machine vision-based industrial inspections, due to changes in conveyor speed or underexposure, the resulting image may produce motion blur, affecting the detection effect. In order to remove the motion blur, a fast linear motion blur method based on RL guided filter was proposed to evaluate the blur parameters of the image first, and the image blurring direction was evaluated using the spectral map integral of the image. Using auto-correlation method to obtain the blur kernel size; then through the proposed RL-guided filtering method to quickly remove the motion blur and suppress the ringing effect, to get a clear image, proved the accuracy of fuzzy kernel estimation Compared with TPKE and HQ, the proposed deblurring method is better for noise robustness and ringing suppression. Through the experiments of real-shot images under different speeds and different workpieces, the objective difference between the objective and the deblurred dimension measurements and the speed of operation were used to prove the superiority of this method in industrial inspection. The results show that the R_PSNR and R_SIMM is superior to the contrast method, which suppresses ringing while maintaining the sharpness of the edges. At the same time, the proposed method has a smaller influence on the measurement of the size and has an advantage over a small-sized workpiece. The maximum error is 0.31 pixels.