Desktop

压缩感知CS——采用小波变换进行稀疏表示，高斯随机矩阵为观测矩阵，重构算法为ROMP算法，对256*256的lena图处理，比较原图和ROMP算法在不同采样比例(0.74、0.5、0.3)下的重构效果，并各运行50次，比较算法性能PSNR和每次的运行时间 (Compressed sensing CS- using wavelet transform as sparse representation, Gaussian random matrix as the observation matrix and ROMP algorithm as the reconstruction algorithm. Compare lena figure and the reconstruction results using ROMP algorithm at different sampling ratio (0.74,0.50.3),then each runs 50 times, compare the performance of PSNR and each running time)

高光谱图像的地物分类技术，对光谱分析并对比检测出地景的算法(Fifth hyperspectral image classification techniques, algorithms for spectral analysis and contrast detect landscapes)

EDLines是一种快速直线检测算子，2012年在ICCV上提出，也是目前处理直线检测最快的算法之一，该算法包含三个步骤：（1）边缘提取：利用Edge Drawing (ED)算法[28,29]从灰度图像中提取边缘片段；（2）线段检测：利用最小二乘法提取直线段；（3）线段确认：遵循Helmholtz定律，从已提取的直线段中摒弃虚假线段。EDLines算法的优越性得益于Edge Drawing (ED)算法能够从灰度图像中准确、快速、稳定地提取出光滑、完整的边缘片段。Edge Drawing (ED)算法包含以下步骤： （1）采用size=5*5,σ=1的高斯核对灰度图像进行平滑滤波，去除噪声； （2）采用一种常用的梯度算子，如Prewitt、Sobel或Scharr等计算平滑后图像中每个像素点的梯度幅度和梯度方向； （3）将梯度图中邻域内幅度最大的像素点标记为锚点，这些锚点成为图像边缘点(edge elements)的概率很大； （4）将相邻的锚点连接成边缘线。从一个锚点起始，ED利用相邻像素的梯度幅度和方向在梯度为最大值的锚点之间游走。 (Edge Drawing (ED) is our recently-proposed, novel, fast edge detection algorithm. What makes ED stand out the existing edge detectors, e.g., Canny, is the following: While the other edge detectors give out a binary edge image as output, where the etected edge pixels are usually independent, discontinuous entities ED produces a set of edge segments, which are clean, contiguous, i.e., connected, chains of edge pixels. Thus, while the output of the other edge detectors requires urther processing to generate potential object boundaries, which may not even be possible or result in inaccuracies ED not only produces perfectly connected object boundaries by default, but it also achieves this in blazing speed compared to other edge detector.)

图像配准。一个遥感图像配准函数。 双线性插值~ (image registration algorithm)

混合像元分解，是定量遥感中非常重要的组成部分之一，在反演地表参数时是必须要考虑的因素之一，文中罗列了分解的一些方法。(mixed pixel decomposition, is the quantification of remote sensing very important component of one of the the inversion of surface parameters is to be considered one of the factors, the paper listed some of the decomposition method.)

BMP,ICO,JPG等图片格式转换，按照例子可以任意格式转换(BMP, ICO, JPG and other image format conversion, according to examples can be converted to any format.)

自己写的Normalized Cut图像分割，聚类程序，程序非常清晰，花了很长时间调试。(Normalize Cut)

说明：  实现DCT与DWT相结合的算法,用matable及小波变换实现所想构思。(The algorithm of combining DCT with DWT is realized, and the idea is realized by matable and wavelet transform.)

图像配准主要是通过两幅图像之间的互相关的最大值来确定配准的位置，从而在另一幅图像中显示配准的图像(Image registration is mainly through the maximum cross-correlation between the two images to determine the location of the registration and registered image, which is displayed in another image)

可以去除不均匀光照、噪声的预处理算法，利用双边滤波和中值滤波的优势。(You can remove the uneven illumination, noise preprocessing algorithm, taking advantage of bilateral filtering and median filtering.)