Impulse_noise_removal

立体视觉代码。主要过程包括：harris角点检测，zncc最优策略匹配，扩散得到更多特征对应，计算disparity和对应颜色值。另外提供了两幅图片以及对应的特征匹配的结果和disparity图像。(Three-dimensional visual codes. The main process includes: harris corner detection, zncc optimal strategy for matching, the proliferation of more features of the corresponding calculation of disparity and the corresponding color value. Also provides a picture and two corresponding feature matching results and disparity images.)

说明：  Hu矩定位图像，可以提取图像的特征，然后进行匹配定位(Hu Moment positioning images, can be extracted image features, and then proceed to match the positioning)

Walsh变换在图像处理是一种非常重要的工具，在图像处理中和radon变换和hough变换都是非常重要的(Walsh transform in image processing is a very important tool, in Image Processing and radon transform and hough transform is very important)

递归与分治演算Gray码，Gray码是一个长度为2N的序列(Recursive partition calculus with the Gray code, Gray code is a sequence of length 2N)

通过非监督分类中的K-均值聚类对Indian pines高光谱图像进行分类(Classification of hyperspectral images of pines Indian by K-means of non supervised classification)

显著图计算算法FT, SR, LC, HC, RC，以及saliency cut 的合集。(an windows msi for install saliency detection software, which includes the implementation for FT, SR, LC, our HC, RC and saliency cut method.)

matlab编写的纹理图像分割 gussian滤波后k-means聚类 并将不同区域用线条表示出来 除m程序还包括测试图片、pdf文件、ppt文件、doc文档(matlab prepared texture image segmentation gussian filtered k-means clustering in different regions and lines that come out with the exception of m program also includes test pictures, pdf file, ppt file, doc file)

矩形检测：用霍夫直线检测方法来对图像中的矩形进行检测。开发环境：vc6，需要安装OpenCV。OpenCV只是用来读取和显示图像等简单操作。(Rectangle detection: Hough line detection method using the image of the rectangle to test. Development Environment: vc6, need to install OpenCV. OpenCV is used to read and display images such as simple operation.)

(cvpr)Matching Local Self-Similarities descriptor自相似描述子((cvpr)Matching Local Self-Similarities across Images and Videos)

图像检测的算法，Faster R-CNN算法，先对整张图像进行卷积计算，然后通过感兴趣区域池化层(RoI Pooling Layer)将选择性搜索算法推荐出来的候选区域和卷积网络计算出的特征映射图进行融合，得到候选区域对应的特征矢量，这种共享卷积计算的操作极大地减少了卷积计算的次数。而且这些特征矢量的维度统一，方便后续的分类工作。通过感兴趣区域池化层处理卷积特征，并将得到的特征送往两个并行计算任务进行训练，分类和定位回归。通过这些方法和改进的框架，Fast R-CNN 用更短的训练和测试时长，取得了比 R-CNN 更好的效果(Faster R-CNN algorithm first convolutes the whole image, then fuses the candidate regions recommended by the selective search algorithm and the feature mapping maps calculated by the convolution network through the RoI Pooling Layer to get the corresponding feature vectors of the candidate regions, which greatly reduces the number of convolution calculations. Moreover, the dimension of these feature vectors is unified, which facilitates the subsequent classification work. The convolution feature is processed by the pooling layer of the region of interest, and the obtained feature is sent to two parallel computing tasks for training, classification and positioning regression. Through these methods and improved framework, Fast R-CNN uses shorter training and testing time and achieves better results than R-CNN.)