computer vision - Is image segmentation required with the SURF algorithm? -

i reading image segmentation, , understood first step in image analysis. read if using surf or sift detect , extract features there no need segmentation. true? there need segmentation if using surf?

the dependency between segmentation , recognition bit more complex. clearly, knowing pixels of image belong object makes recognition easier. however, relationship works in other direction: knowing in image makes easier segmentation. however, simplicity, speak simple pipeline segmentation performed first (for instance based on simple color model) , each of segments processed.

your question asks surf features. however, in context, important surf local descriptor, i.e. describes small image patches around detected keypoints. keypoints should points in image information relevant recognition problem can found (interesting parts of image), points can reliably detected in repeatable fashion on images of objects belonging class of interest. result, local descriptor cares pixels around points selected keypoint detector , each such keypoint extracts small feature vector. on other hand global descriptor consider pixels within area, typically segment, or whole image.

therefore, perform recognition in image using global descriptor, need first select area (segment) want features extracted. these features used recognize content of segment. situation bit different local descriptor, since describes local patches keypoint detector determines relevant. result, multiple feature vectors multiple points in image, if not perform segmentation. each of these feature vectors tells content of image , can try assign each such local feature vector "class" , gather statistics understand content of image. such simple model called bag-of-words model.