SIFT: Computer Vision Classics

In object recognition problems, we must identify features that do not change with the image scale, rotation,

3D viewpoint changes,

illumination, and noise. Moreover, these features must be substantially different from each other. This helps us to match the features against a database of features without ambiguity. We must be able to locate the features in both spatial (position) and frequency(scale and details) domains. We will use a step-wise (cascade) filtering technique. This avoids going

...