不均匀照明下采集图像的校正方法及硬件加速

Correction Method of Image Under Nonuniform Illumination and Its Hardware Acceleration

在表面缺陷检测的过程中,需要通过预处理增强目标缺陷与背景之间的对比,再进行缺陷与背景的区域分割。而当探测光源在目标表面产生的光照不均匀时,需要先对采集图像进行光照均匀性校正,否则较难进行之后的缺陷目标提取。本文提出了一种基于小波变换和动态阈值的校正方法,适用于不均匀照明下缺陷图像的增强,可以很好地进行照明分量的均衡。与已有方法相比,该方法在光照的均衡效果、普适性、直观性、后续目标区域的提取等方便有显著优势,但是计算时间相对较长。针对该问题,本文使用VivadoHLS设计了IP加速核,将运算效率提高了20倍,可广泛运用在不均匀照明下表面缺陷的识别。

In the process of surface defect detection, it is necessary to enhance the contrast between the target defect and the background through image preprocessing, and then divide the target and background regions. When the illumination generated by the light source on the target surface is nonuniform, the image should go through uniformity correction processing, otherwise it would be difficult to perform subsequent target region extraction from the background. In this paper, a correction method based on wavelet transform and dynamic threshold is proposed, which is suitable for the image enhancement under nonuniform illumination and can significantly balance the illumination component. Compared with other methods, the proposed method has significant advantages on illumination equilibrium effects, universality, visuality, and subsequent extraction of target regions, but the calculation time is longer. In response to this problem, this paper uses Vivado HLS to design an IP acceleration core, which improves the computational efficiency by 20 times, and can be widely used for the identification of surface defects under nonuniform illumination.