自然图像的无参考模糊检测与局部模糊区域分割

发布时间：2018-10-22 18:04

【摘要】：针对自然图像的模糊强度检测和局部模糊区域分割,提出一种无参考无训练的检测分割算法.首先对待测图像进行再模糊;然后对再模糊图像和待测图像逐点进行小邻域离散余弦变换,得到待测图像的模糊强度分布;最后结合K-Means聚类算法和形态学运算对图像的局部模糊区域进行分割提取.实验结果表明,采用文中算法得到的模糊强度分布图能够有效地检测和分割图像的清晰区域与模糊区域;与同类算法相比,对于不同模糊形式和不同复杂度的图像,该算法在查准率、查全率和F值等图像分割性能指标上表现较为优异,与人眼主观分割结果具有较高一致性,且该算法无需进行数据训练,具有较高的时间效率.
[Abstract]:Aiming at the fuzzy intensity detection and local fuzzy region segmentation of natural images, a non-reference and no training detection and segmentation algorithm is proposed. Firstly, the image is re-blurred, then the small neighborhood discrete cosine transform is carried out on the re-blurred image and the image to be tested, and the fuzzy intensity distribution of the image under test is obtained. Finally, K-Means clustering algorithm and morphological operation are used to segment and extract the local fuzzy region of the image. Experimental results show that the fuzzy intensity distribution map obtained by the proposed algorithm can effectively detect and segment the clear region and the fuzzy region of the image, and compare with the similar algorithms, for the images with different fuzzy forms and different complexity, the fuzzy intensity distribution map can effectively detect and segment the clear region and the fuzzy region of the image. The algorithm has excellent performance in precision, recall and F-value, and has high consistency with subjective segmentation results of human eyes. Moreover, the algorithm does not need data training and has high time efficiency.
【作者单位】：中国科学技术大学精密机械与精密仪器系;石家庄铁道大学机械工程学院;中国科学技术大学国家同步辐射实验室;
【基金】：国家自然科学基金联合基金(U1332130) 高等学校学科创新引智计划(B07033) 国家“九七三”重点基础研究发展计划项目(2014CB931804)
【分类号】：TP391.41

【相似文献】