Detecting Corners using susan corner detection algorithm
Smallest Univalue Segment Assimilating Nucleus (SUSAN): Smith and Brady (IJCV, Vol. 23(1), pp. 45–78, 1997) presented an entirely different approach to the 1D and 2D feature detection in images, such as edges and corners, respectively. The SUSAN corner detection algorithm does not require derivative operation; hence, it can work well with noisy images. The main idea of the SUSAN is the use of a mask to count the num- ber of pixels having the same brightness as the center pixel. By comparing the number of pixel having the same brightness as the center pixel with a threshold, the detector can determine whether the center pixel is a corner.
Circular mask M consisting of 37 pixels should be used. The central pixel of the mask is called a nucleus. Then intensities of all pixels within a mask are compared with an intensity of a nucleus and an area of “similar” pixels is marked. This area is called USAN (Univalue Segment Assimilating Nucleus) and it conveys the most important information on a local structure of an image. Analyzing the size, centroid and the second moments of USAN the exact information on a type of local structure around a nucleus is inferred, such as edges or corners. For those regions, inverted USAN area shows strong peaks –thus the term SUSAN – i.e. the smallest USAN. This approach has an additional advantage of not using any derivatives which are cumbersome to use in the presence of noise.
Place a circular mask around a pixel (i.e., nucleus) • Calculate the number n(r 0 ) of pixels within the circular mask which have similar brightness to the nucleus in accordance with Equation 3.1. Such pixels constitute the USAN. • Compute the strength of a corner from Equation 3.2. • Use non-max suppression to find corners.