KmeansPP

这是基于张亶，陈刚《偏微分方程的图像处理》书的实验及程序(image processing )

用块匹配法求超分辨率复原中的运动矩阵!提出了一种快速，半像素的三步法！(Block-matching method using super-resolution recovery of movement in the matrix! Presents a fast, three-step half-pixel!)

利用matlab实现图像处理里的实现图像边缘检测、图像分割、特征提取以及图像数字识别等任务。 对图像分别进行模糊、锐化、和直方图均衡化处理，画出处理结果。实现整幅图像的边缘检测，可以看出有肖像画或素描的效果，画出处理结果。(Matlab realize image edge detection in image processing, image segmentation, feature extraction and image numeral recognition task. Fuzzy image sharpness, and histogram equalization paint treatment results. The whole image edge detection, portrait or sketch effect can be seen the paint treatment results.)

图像峰度计算函数，可作为图像质量评价的一个标准(Kurtosis calculated image function, image quality evaluation can be used as a standard)

semi-supervised learning 半监督局部线性嵌入(semi-supervised learning )

能计算一幅噪声图像的加权峰值信噪比，以评价各种图像去噪方法的效果(Can calculate an image of the weighted noise PSNR to evaluate the effect of image denoising methods)

多光谱图像出理，离散傅立叶变换，低通滤波，TM条带影像去除(multispectral image integration of remote sensing)

本程序应用相位一致性进行边缘检测，较以往应用灰度信息进行边缘检测算法，该算法与光照无关，且抗噪声性能强(phase of the application process consistency edge detection, gray than in the past Application information edge detection algorithm, The algorithm has nothing to do with illumination, and strong anti-noise performance)

实现图像的几何变换，包括图像的平移、镜像变换、缩放、转置以及旋转。(Achieve geometric transformation of image, including the image translation, image transformation, scaling, transpose and rotation.)

混合高斯(Mixture of Gaussian, MOG) 背景建模算法和Codebook 背景建模算法被广泛应用于监控视频的运动目标检测问题, 但混合高斯的球体模型通常假设RGB 三个分量是独立的, Codebook 的圆柱体模型假设背景像素值在圆柱体内均匀分布且背景亮度值变化方向指向坐标原点, 这些假设使得模型对背景的描述能力下降. 本文提出了一种椭球体背景模型, 该模型克服了混合高斯球体模型和Codebook 圆柱体模型假设的局限性, 同时利用主成分分析(Principal components analysis, PCA) 方法来刻画椭球体背景模型, 提出了一种基于主成分分析的Codebook 背景建模算法. 实验表明, 本文算法不仅能够更准确地描述背景像素值在RGB 空间中的分布特征, 而且具有良好的鲁棒性.(The background modeling algorithm of mixture of Gaussian (MOG) and codebook is widely used in moving object detection of surveillance video. However, the ball model of MOG usually assumes that the three components of RGB are independent, while the cylinder model of codebook assumes that the value of background pixel is distributed uniformly within the cylinder and the changing direction of brightness points to the origin of the coordinate system. These assumptions reduce the descriptive capability for background modeling. Therefore, the paper proposes an ellipsoid- based background model, which overcomes the MOG and codebook0s limitations. By using principal component analysis to depict the ellipsoid background model, a novel PCA-based codebook background modeling algorithm is proposed. Experiments show that this algorithm can not only give more accurate description of the distribution of background pixels but also have a better robustness.)