MSR

Downsample image console program using opencv.

Compression of face images impact the performance of face recognition (FR) systems. JPEG Region of Interest (JROI) compression maintains high image quality in facial regions while compressing the background more, with minimal impact on FR performance.(Compression of face images impact the perf ormance of face recognition (FR) systems. JPEG Region of Interest (JROI) compression maintai ns high image quality in facial regions while co mpressing the background more, with minimal impact on FR performance.)

基于深度学习的图像超分辨率算法。 参考论文 Learning a Deep Convolutional Network for Image Super-Resolution (ECCV 2014) (Super resolution based on deep learning Refer to Learning a Deep Convolutional Network for Image Super-Resolution (ECCV 2014))

在人脸数据库中应用PCA，以及重建误差分析(Applying PCA in human face dataset and doing analysis on reconstruction differences )

本文讲述了一种基于图像颜色直方图的图像相似度的度量的方法,。并用Matlab实现，有源代码(This article describes a method based on the image color histogram image similarity measure. And using Matlab, source code)

Source code to segment skin colour.

采用超像素进行分割图像，利用到了超像素和Ncut的方法，采用了c与maltab混合编程。(Code to compute superpixels. use cncut and superpixel code for segmentation. Mex program c and matlab)

做项目时写的关于相机图像采集、控制、图像处理、坐标计算、坐标显示等功能的小软件(Do item written on the camera image acquisition, control, image processing, coordinates of the calculation, the coordinates show features such as small software)

给定一幅图像，输出其直方图。给定一个线性变换函数(可以输入不同的参数)，实现图像的灰度拉伸；实现对图像的均衡化。 实验测试图像：任意 (histogram；Gray stretch；Image equalization)

提出了一种基于非下采样Contourlet 变换的自适应图像去噪方法 首先对噪声图像进行非下采样Contourlet 变换，得到各 个尺度各个方向子带的系数，再根据该系数的能量自适应地调整去噪阈值 实验表明，与Contourlet 多尺度阈值去噪 Contourlet 自 适应阈值去噪相比，该方法在保留图像边缘细节的同时，能提高图像的PSNR 值，减少了Gibbs 现象(A new adaptive method of image de-noising based on NonSubsampled Contourlet Transform （NSCT）is presented.Firstly， the noisy image is decomposed into a set of multiscale and multidirectional frequency subbands by NSCT， and then the threshold is adapted automatically according to the energy of subband coefficients.Compared with the multi-scale threshold using contourlet transform and using adaptive threshold based on contourlet transform， the simulation results show that the performance of this method is superior in PSNR， meanwhile it can maintain the image edge and reduce the Gibbs phenomen)