beijing

说明：  一、主要研究内容 1.基本内容 本设计目标在于利用Matlab强大的图像处理能力和实用便捷的编程方法，通过处理包含人脸的视频帧系列图像，灰度积分投影技术的眼睛定位方法，进而利用perclos计数，计算眨眼率，从而得到比较准确的疲劳状况。包括lun文初稿撰写。 2.具体要求 本设计基于灰度积分投影技术的眼睛定位方法，再结合perclos技术。首先通过图像预处理技术得到灰度分配较为均匀的图像,然后分别利用水平和垂直灰度积分投影曲线结合人脸的结构特征找到眼睛的位置坐标,实现了准确的眼睛定位,通过perclos技术技术眨眼率，根据先验值得到是否疲劳。 二、算法原理 视频输入——预处理去除非人脸区域——人脸定位——人眼定位（灰度积分投影）——眼睛张合度——perclos技术统计——输出结果 三、GUI界面设计 [需要有一定编程功底](1. Basic contents The goal of this design is to use the powerful image processing ability of MATLAB and the practical and convenient programming method, through processing the video frame series images including face, the eye location method of gray integral projection technology, and then using PERCLOS count to calculate the blink rate, so as to get a more accurate fatigue condition. Including the first draft of Lun.)

本文献讲述了使用多维的结构张量进行基于分水岭的图像增强和去噪方法(The literature describes the use of multi-dimensional structure of tensor-based watershed image enhancement and denoising)

现有自然图像素描画效果生成算法存在容易造成画面细节丢失，笔画线条单一等问题。针对这些问题，本文提出了一种改进的自然图像素描画效果生成算法，通过模拟艺术家创作素描画过程，分步生成素描画图像底纹和素描画场景笔画轮廓，采用8个方向卷积滤波提取图像中不同方向的细节信息，并使用不同长度不同粗细的笔画来模拟艺术家在创作过程中不同力度的绘画效果，最后合并素描画笔画效果和素描画图像底纹，得到与原输入自然图像内容一致，色调与艺术家绘制的素描画效果相同，笔画和底纹形式与实际素描画相似，真实感较强的素描画效果输出图像。大量实验结果表明本文方法可以生成高质量的自然图像素描画效果，与现有成熟的铅笔画生成技术相笔，本文方法具有较强的可比性。(The existing algorithms for natural image sketch production are easy to cause some problems such as the loss of image details, singleness of the stroke line. To solve these problems, this paper proposes an improved algorithm for the natural image sketch effect producing. By simulating the artist creation process of sketch, step-by-step generate image shading sketch image shading and drawing the scene outline stroke of the sketch image, using 8 directions of convolution filter to extract the details of different directions in the image, furthermore, using the stroke of different length and thickness to simulate the painting effect of different strength in the process of artist creation, finally we merge the sketch stroke effect and sketch image background to get the output image, which have consistent content with the original input image, have the same tonal effect with artists draw sketch, have the similar Stroke and background form with the actual sketch and have strong real sense in)

图像拼接小程序，使用直接对比的方法进行拼接并用加权平均的方法进行融合 (a simple program for image splicing )

这是一个matlab的源码。代码实现了Nonrigid Registration Using Free-Form Deformation.(It is a matlab source code. It implements "Nonrigid Registration Using Free-Form Deformation".)

使用tensorflow框架搭建神经网络，从单目图像中估计深度值(Using tensorflow framework to build a neural network to estimate depth from monocular images)

Google Doodle推出了一款极其炫酷的LOGO，整个LOGO使用了CSS Sprite技术，利用一个初始图片和一张画满各个动作的拼接图片，实现了动画，而非传统的GIF动画图像(Google Doodle launched a very cool LOGO, LOGO using CSS Sprite entire technology, the use of an initial picture and a painting full picture stitching each action to achieve the animation, instead of the traditional animated GIF images)

orthogonal Matching Pursuit(OMP),CoSaMP算法的Matlab实现。(orthogonal Matching Pursuit(OMP),CoSaMP Algorithms)

泊松图像融合算法的实现，这是我自己写的，很多地方可以改进，希望对学习这方面的有帮助(poisson image editing)

说明：  性能上比经典分水岭有很大的提高。特别对于自然图像的分割，效果特别好。(Performance than the classical watershed have greatly improved. Especially for natural image segmentation, the effect particularly good.)