最大均值差异(Maximum mean discrepancy,MMD)作为一种能有效度量源域和目标域分布差异的标准已被成功运用.然而,MMD作为一种全局度量方法一定程度上反映的是区域之间全局分布和全局结构上的差异.为此,本文通过引入局部加权均值的方法...最大均值差异(Maximum mean discrepancy,MMD)作为一种能有效度量源域和目标域分布差异的标准已被成功运用.然而,MMD作为一种全局度量方法一定程度上反映的是区域之间全局分布和全局结构上的差异.为此,本文通过引入局部加权均值的方法和理论到MMD中,提出一种具有局部保持能力的投影最大局部加权均值差异(Projected maximum local weighted mean discrepancy,PMLWD)度量,结合传统的学习理论提出基于局部加权均值的领域适应学习框架(Local weighted mean based domain adaptation learning framework,LDAF),在LDAF框架下,衍生出两种领域适应学习方法:LDAF MLC和LDAF SVM.最后,通过测试人工数据集、高维文本数据集和人脸数据集来表明LDAF比其他领域适应学习方法更具优势.展开更多
针对灰度图像受脉冲噪声污染后的恢复处理问题,提出了一种改进的自适应中值滤波算法。该方法根据脉冲噪声的分布特点,采用极大值、极小值和领域均值判定准则进行噪声点的检测,然后用检测窗口内最小非噪声点集合的中值作为噪声点的滤波...针对灰度图像受脉冲噪声污染后的恢复处理问题,提出了一种改进的自适应中值滤波算法。该方法根据脉冲噪声的分布特点,采用极大值、极小值和领域均值判定准则进行噪声点的检测,然后用检测窗口内最小非噪声点集合的中值作为噪声点的滤波输出。实验结果表明,与其他几种算法相比,文中算法不仅在峰值信噪比(Peak Signal to Noise Ratio)和结构相似度(Structural Similarity,SSIM)上有较大优势,而且还具有较低的时间复杂度和更好的自适应性。也进一步说明该方法不仅能有效地检测并滤除噪声点,还能较好地保护图像的边缘细节。展开更多
In this paper, a flexible high-precision calibration method suitable for industrial field was proposed. The complexity of the coordinate transformation was simplified by choosing the camera coordinate system as the un...In this paper, a flexible high-precision calibration method suitable for industrial field was proposed. The complexity of the coordinate transformation was simplified by choosing the camera coordinate system as the unified reference coordinate system. A flexible planar calibration pattern was introduced to the calibration process, which can be arbitrarily placed and from which the known feature points can be extracted to construct other unknown feature points. With the known intrinsic parameters, the laser projector plane equation was fitted by the multi-noncollinear points, which were acquired through the principle of triangulation and the projective invariance of cross ratio. With this method, the strict alignment and multiple times of coordinate transformation can be avoided. Experimental results showed that the arithmetic mean of the root mean square(RMS) error of distance was 0.000 7 mm.展开更多
文摘最大均值差异(Maximum mean discrepancy,MMD)作为一种能有效度量源域和目标域分布差异的标准已被成功运用.然而,MMD作为一种全局度量方法一定程度上反映的是区域之间全局分布和全局结构上的差异.为此,本文通过引入局部加权均值的方法和理论到MMD中,提出一种具有局部保持能力的投影最大局部加权均值差异(Projected maximum local weighted mean discrepancy,PMLWD)度量,结合传统的学习理论提出基于局部加权均值的领域适应学习框架(Local weighted mean based domain adaptation learning framework,LDAF),在LDAF框架下,衍生出两种领域适应学习方法:LDAF MLC和LDAF SVM.最后,通过测试人工数据集、高维文本数据集和人脸数据集来表明LDAF比其他领域适应学习方法更具优势.
文摘针对灰度图像受脉冲噪声污染后的恢复处理问题,提出了一种改进的自适应中值滤波算法。该方法根据脉冲噪声的分布特点,采用极大值、极小值和领域均值判定准则进行噪声点的检测,然后用检测窗口内最小非噪声点集合的中值作为噪声点的滤波输出。实验结果表明,与其他几种算法相比,文中算法不仅在峰值信噪比(Peak Signal to Noise Ratio)和结构相似度(Structural Similarity,SSIM)上有较大优势,而且还具有较低的时间复杂度和更好的自适应性。也进一步说明该方法不仅能有效地检测并滤除噪声点,还能较好地保护图像的边缘细节。
基金Supported by the National Natural Science Foundation of China(No.51105273)
文摘In this paper, a flexible high-precision calibration method suitable for industrial field was proposed. The complexity of the coordinate transformation was simplified by choosing the camera coordinate system as the unified reference coordinate system. A flexible planar calibration pattern was introduced to the calibration process, which can be arbitrarily placed and from which the known feature points can be extracted to construct other unknown feature points. With the known intrinsic parameters, the laser projector plane equation was fitted by the multi-noncollinear points, which were acquired through the principle of triangulation and the projective invariance of cross ratio. With this method, the strict alignment and multiple times of coordinate transformation can be avoided. Experimental results showed that the arithmetic mean of the root mean square(RMS) error of distance was 0.000 7 mm.
