Point features can be used to ﬁnd a sparse set of corresponding locations in different images.
There are two main approaches to ﬁnding feature points and their correspondences:
- The ﬁrst is to ﬁnd features in one image that can be accurately tracked using a local search technique such as correlation or least squares; -more suitable when images are taken from nearby viewpoints or in rapid succession (e.g., video sequences)
- The second is to independently detect features in all the images under consideration and then match features based on their local appearance – more suitable when a large amount of motion or appearance change is expected, e.g., in stitching together panoramas
keypoint detection and matching pipeline four separate stages:
- feature detection (extraction) stage – each image is searched for locations that are likely to match well in other images
- second feature description stage – each region around detected keypoint locations in converted into a more compact and stable (invariant) descriptor that can be matched against other descriptors
- third feature matching stage – efﬁciently searches for likely matching candidates in other images
- fourth feature tracking stage – an alternative to the third stage that only searches a small neighborhood around each detected feature and is therefore more suitable for video processing
- point detection – finds points or patches in the image that have saliency – interest points
- point description – assigns a feature vector to each point
The NNs of a point in one image are found in another image:
There are many existing point detection and description algorithms, SIFT return not only the location of the points, but also the scale and orientation.
Point matching as classification – instead of computing point feature vectors and finding the NNs, each point in the ‘training’ image is a class.
– junctions of the contours
– more stable features over the changes of viewpoint
– large variations in the neighbourhood of the point in all directions
– Corner is the good feature to match!
Looking at intensity values within a small window, shifting the window in any direction should yield a large change in appearance.
Harris Corner detector
Harris corner detector gives a mathematical approach for determining which case holds:
- flat region: no change in all directions
- edge: no change along the edge direction
- corner: significant change in all directions