CAJ | 학술논문

为便于集成电路（IC）真实缺陷形貌图的缺陷特征提取，提出了一种结合直方图均衡化（HE）和多尺度 Retinex 彩色恢复（MSRCR）算法的彩色图像增强新算法。用直方图均衡化对彩色图像进行增强，可以显著提高对比度，但会降低原图的信息熵；用 Retinex 算法对彩色图像进行增强，可以显著提高暗区域的细节，但会产生泛白、颜色失真和对比度低的现象。新算法根据两种算法处理结果的特点，将图像先分别进行HE 增强和 MSRCR 增强，然后按照一定的图像融合规则进行加权融合，经过大量的测试统计，得到了一个最佳权重。实验证明，改进的算法使图像的亮度、对比度、细节等都有很大的增强，不仅改善了图像的整体视觉效果，而且得到了最大的信息熵，能更好地刻画 IC 缺陷细节，有利于后续的目标检测和缺陷特征提取，并验证了算法的通用性。
위편우집성전로（IC）진실결함형모도적결함특정제취，제출료일충결합직방도균형화（HE）화다척도 Retinex 채색회복（MSRCR）산법적채색도상증강신산법。용직방도균형화대채색도상진행증강，가이현저제고대비도，단회강저원도적신식적；용 Retinex 산법대채색도상진행증강，가이현저제고암구역적세절，단회산생범백、안색실진화대비도저적현상。신산법근거량충산법처리결과적특점，장도상선분별진행HE 증강화 MSRCR 증강，연후안조일정적도상융합규칙진행가권융합，경과대량적측시통계，득도료일개최가권중。실험증명，개진적산법사도상적량도、대비도、세절등도유흔대적증강，불부개선료도상적정체시각효과，이차득도료최대적신식적，능경호지각화 IC 결함세절，유리우후속적목표검측화결함특정제취，병험증료산법적통용성。
In order to conveniently extract the extra material defect features from an IC real image , this paper proposes a new method for a color image enhancement combined with histogram equalization ( HE) and Multi-Scale Retinex with Color Restoration ( MSRCR) . Using histogram equalization to color image enhancement can significantly improve the contrast but will reduce the original information entropy , the Retinex algorithm can improve the details of the dark area but will lead to the phenomena such as the white and color distortion , low contrast . The new algorithm , according to the characteristics of the processing results of the above two algorithms , weightily fusing the HE enhanced image and the MSRCR enhanced image , has been one of the best weighting factors after a lot of test statistics . Experimental results show that the improved algorithm produces greater enhancement in the image's brightness , contrast , detail , and others and that it not only improves the overall visual effect of the image , but also gives the maximum information entropy . Through objective and subjective evaluation , it is shown that the algorithm has a fantastic effect on enhancement of color image , compared to the HE and MSRCR algorithm that process separately , and that it can better describe IC defects in detail , which is conducive to the detection and defect feature extraction of the subsequent target , and verify the versatility of the algorithm .