sift算法在cv领域的重要性不言而喻，该作者的文章引用率在cv界是number1.本篇博客只是本人把sift算法知识点整理了下，以免忘记。本文比较早的一篇博文opencv源码解析之(3)：特征点检查前言1 中有使用opencv自带的sift做了个简单的实验，而这次主要是利用Rob Hess的sift源码来做实验，其实现在的opencv版本中带的sift算法也是Rob Hess的，只是稍微包装了下。
下面来做下试验,试验sift代码采用Rob Hess的代码，opencv目前版本中的sift源码也是采用Rob Hess的。代码可以在他的主页上下载：http://blogs.oregonstate.edu/hess/code/sift/
这里我下载的是windows版本的，并采用Qt做了个简单的界面。
sift algorithm is self-evident the importance of the cv field, the author of the article referenced in the cv sector is number1 this blog I just sift algorithm knowledge compiled under, so as not to forget. This article compares the earlier blog post opencv source code analysis (3): check the preface of the feature points 1 the opencv bring their own sift to do a simple experiment, but this time Rob Hess' s sift source to experiment. In fact, with the opencv version sift algorithm is Rob Hess, only slightly packing the next. Below to do the next test, test sift code using the code of Rob Hess, sift source opencv version of Rob Hess. The code can be downloaded on his home page: http://blogs.oregonstate.edu/hess/code/sift/ here I downloaded the windows version, and using Qt to do a simple interface. (2012-08-16, QT, 1061KB, 下载517次)
本篇論文提出的整合式解交錯(Integrated De-interlacing)的演算法，可以有效提昇移
動區域的畫面，但是當移動估計不正確時，反而會使移動補償後的畫面變得很差，為了
改善這種情況，因此結合移動可適性解交錯的優點，並將空間圖場內插(Spatial
Interpolation)的方式改成ELA(Edge Line Average)來設計，經過電腦模擬的結果發現，不僅在視覺上提高畫面的解析度，在某些影像峰值訊號雜訊比(Peak Signal Noise Ratio ,
PSNR)也比線平均解交 錯(Line Average De-interlacing)多出好幾分貝的畫質增益。
此外，在整合式解交錯演算法中也增加影片偵測(Film Detection)和影像加強(Image
Enhancement)的演算法設計，在這樣演算法的架構下，透過影片偵測的演算法，我們可
真實地還原3:2 Pull Down 的影片格式，而不會有鋸齒狀(Saw-Toothed)的畫面出現，而影
像加強的演算法，則可以在解交錯後，經過影像的調整，使輸出畫面呈現不同的效果，
達到消費者的需求。
The main theme of this thesis is an integrated de-interlacing system, which incorporates
several known and improved techniques in a nice manner to produce good de-interlaced image
quality. We first develop an accurate motion detector that classifies image regions into
stationary, low-motion, and high-motion categories. The simple field merging method is
applied to the stationary regions. The edge line average interpolation method is applied to the
slow-motion regions. Finally, the motion-compensated interpolation is applied to the
high-motion regions. In addition, hierarchical motion estimation and motion vector smoothing
techniques are employed to enhance the quality of estimated motion vectors. Our computer
simulation shows that the subjective image quality is improved by using the proposed scheme.
Also, its PSNR measures are better than the conventional spatial or temporal interpolation
schemes. (2010-10-26, PDF, 1144KB, 下载10次)