东南大学学报(自然科学版)
東南大學學報(自然科學版)
동남대학학보(자연과학판)
JOURNAL OF SOUTHEAST UNIVERSITY
2014年
4期
735-739
,共5页
中值滤波%微小缺陷分割%缺陷检测%特征提取
中值濾波%微小缺陷分割%缺陷檢測%特徵提取
중치려파%미소결함분할%결함검측%특정제취
median filter%small defect segmentation%defect detection%feature extraction
针对微小缺陷在复杂背景图像情形下分割难的问题,提出了一种基于像元搜索算法的微小缺陷检测方法。首先采用直方图均衡化提升背景与缺陷目标的对比度,在分析噪声分布特点的基础上,利用基于中值和均值滤波的改进滤波算法对图像进行去噪等前期预处理;然后根据背景灰度分布,在目标分割过程中采用分块、按方差大小排除背景图像块、初定目标和剔除伪目标的缺陷像元搜索算法;最后采用矩形度和区域占空比进行缺陷特征提取。结果表明,对于背景不均匀、目标与背景区分不明显这类复杂背景图像,所提出算法相对于传统的Otsu等算法能够更好地分割出弱小缺陷目标,提高了检测缺陷的准确性。
針對微小缺陷在複雜揹景圖像情形下分割難的問題,提齣瞭一種基于像元搜索算法的微小缺陷檢測方法。首先採用直方圖均衡化提升揹景與缺陷目標的對比度,在分析譟聲分佈特點的基礎上,利用基于中值和均值濾波的改進濾波算法對圖像進行去譟等前期預處理;然後根據揹景灰度分佈,在目標分割過程中採用分塊、按方差大小排除揹景圖像塊、初定目標和剔除偽目標的缺陷像元搜索算法;最後採用矩形度和區域佔空比進行缺陷特徵提取。結果錶明,對于揹景不均勻、目標與揹景區分不明顯這類複雜揹景圖像,所提齣算法相對于傳統的Otsu等算法能夠更好地分割齣弱小缺陷目標,提高瞭檢測缺陷的準確性。
침대미소결함재복잡배경도상정형하분할난적문제,제출료일충기우상원수색산법적미소결함검측방법。수선채용직방도균형화제승배경여결함목표적대비도,재분석조성분포특점적기출상,이용기우중치화균치려파적개진려파산법대도상진행거조등전기예처리;연후근거배경회도분포,재목표분할과정중채용분괴、안방차대소배제배경도상괴、초정목표화척제위목표적결함상원수색산법;최후채용구형도화구역점공비진행결함특정제취。결과표명,대우배경불균균、목표여배경구분불명현저류복잡배경도상,소제출산법상대우전통적Otsu등산법능구경호지분할출약소결함목표,제고료검측결함적준학성。
In order to solve the difficulty of segmenting the small defects from images with a com-plex background, a new detection method based on the pixel search is proposed.Firstly, histogram equalization is used to enhance the contrast between the target and the background, and an improved filtering algorithm based on median and mean filtering is applied to denoising according to the char-acteristics of the noise distribution.Then, according to the background gray distribution, a new seg-ment technique based on the pixel search is proposed, which includes four steps:division of the im-age, exclusion of the background blocks based on the variance, determination of the preliminary de-fects target and elimination of the fake targets.Finally, the rectangularity and the regional duty ratio are used to extract the defect features.The experimental results show that compared with the tradi-tional Otsu algorithm, this segmenting algorithm can more successfully separate the small defects tar-get from images with complicated background and overlap regions between the target and back-ground, which can improve the accuracy of defect detection.