ROA

图像分解算法，四种方法可选，包括threshold, K means threshold, region growing以及gradient based segmentation(Image decomposition algorithm, four methods available, including the threshold, K means threshold, region growing and gradient based segmentation)

一般自然图像的梯度分布符合重尾分布，重尾分布也就是超拉普拉斯分布，根据这个特点进行图像复原。(The gradient of a natural image in line with heavy-tailed distribution, which is ultra-heavy-tailed distribution Laplace distribution, image restoration based on this feature.)

说明：  几种图像融合评价指标，AG,EN,FMI,MI,Qab,Qw,SF,SSIM，即平均梯度AG、信息熵EN、特征互信息FMI、互信息MI、边缘信息保持量Qab、加权融合质量评价因子Qw、空间频率SF和相似性指数SSIM。能较好的描述图像融合结果，稳定性较好。(Several evaluation indexes of image fusion, Ag, en, FMI, MI, qAB, QW, SF, SSIM, namely average gradient AG, information entropy En, feature mutual information FMI, mutual information MI, edge information retention qAB, weighted fusion quality evaluation factor QW, spatial frequency SF and similarity index SSIM. It can describe the result of image fusion well and has good stability.)

2015年最新国外算法，基于加权视觉词典的人群密度估计(crowd entimatin by a weighted bag of visual word model)

这是一个多重网格的程序，可以对偏微分方程进行计算！(This is a multi-grid procedure, partial differential equations can be calculated!)

Real-time Specular Highlight Removal实现论文“Real-time Specular Highlight Removal Using Bilateral Filtering”中算法(Real-time Specular Highlight Removal Papers " Real-time Specular Highlight Removal Using Bilateral Filtering" in the algorithm)

使用双线性内插或二次曲面内插放大图像，然后使用彩色自蛇模型去除边缘模糊和锯齿效应(Using bilinear interpolation or quadratic surface interpolation to enlarge the image, then use the color removed from the snake model and the sawtooth effect of edge blur)

实现了Chan－Vese算法，参考文献active contour without edges, 图像处理方面引用率最高的文献之一(Realize the Chan-Vese algorithm, reference active contour without edges, image processing, to quote one of the highest rates of the literature)

四个IMAGE窗口，分别对图像进行读取，最大值滤波，最小值滤波及中值滤波。(Four IMAGE window, read the image, respectively, maximum filter, the minimum filter and median filter.)

提出了一种结合USFFT Curvelet 变换的各向异性扩散图像去噪模型。它有机结合了Curvelet 变换和各向 异性扩散（P-M扩散）两者的优点。通过P-范数方法选择合适的梯度阈值K，P-M扩散过程通过处理经过Curvelet 变换得到的图像的不同尺度的Curvelet 系数矩阵，实现了建立在对图像多尺度分析的基础上的新P-M扩散模型。 实验表明，新模型的处理结果能有效避免传统P-M 扩散出现的阶梯效应，同时更好地保留图像的纹理和细节。 关键词：图像去噪算法；各向异性扩散；Curvelet 变换；P-M 扩散(Abstract: An image de-noising model by integrating anisotropic diffusion with USFFT Curvelet transform was proposed, which combined the strongpoint of Curvelet transform with anisotropic diffusion (P-M diffusion). By choosing appropriate gradient threshold K through p-norms and carrying out the P-M diffusion process depend on the different scale matrixes of Curvelet coefficient of the image from Curvelet transform iterations, as a result, the improved model made it possible to carry out the new P-M diffusion de-noising process based on multi-scale analysis of the image. The experiment results have demonstrated that the model can avert the stair-casing effect in the traditional P-M diffusion effectively and keep the textures and details of images better.)