SURF uses an approximated DoG and the integral image (very similar to the method used in the famous Viola and Jones’ adaboost face detector) trick: An integral image is just an image which its each pixel value is the sum of all the original pixel values left and above it. The advantage of integral image is that it can compute block subtraction between any 2 blocks with just 6 calculations, so finding SURF features could be several order faster than the traditional SIFT features.
ORB 用FAST作为特征点提取的算法，更快了，添加了特征点的主方向，这样就具有了旋转不变性。采用贪婪穷举的方法得到了相关性较低的随机点对,受噪声的影响很大. (ORB 选择该像素为中心的一个小patch作为比较对象，提高了抗噪能力。) BRIEF 则是对原图像滤波，降低噪声的影响
- Most of the vision work take a high dimension data and turn it into a lower dimension data while only throwing away uninformative data, with the aim of classifying it easier. SIFT points, HOG, SURF are just doing this – Trying to find the lower dimension data that tells most.
- And then we head to the second step to classify these lower dimension data. It could be as simple as nearest neighbor, probability compared with your trained model or any machine learning algorithms such as Adaboost, SVM, neural network, etc. This step classifies all these points into different categories.
Changing the parameters just slightly change the sub space your images are mapped to, or throwing these points into bins slightly differently. If it only works when you tweak it a lot, you are probably mapping to the wrong space or throwing points the wrong way.