The Radon Transform

Background

In recent years the Hough transform and the related Radon transformhave received much attention. These two transforms are able to transformtwo dimensional images with lines into a domain of possible line parameters,where each line in the image will give a peak positioned at the correspondingline parameters. This have lead to many line detection applications withinimage processing, computer vision, and seismics.

Several definitionsof the Radon transform exists, but the are related, and a very popularform expresses lines in the form rho=x*cos(theta)+y*sin(theta), where thetais the angle and rho the smallest distance to the origin of the coordinatesystem. As shown in the two foloowing definitions (which areidentical), the Radon transform for a set of parameters (rho,theta) isthe line integral through the image g(x,y), where the line ispositioned corresponding to the value of (rho,theta). The delta() isthe Dirac delta function which is infinite for argument 0 and zero forall other arguments (it integrates to one), and in digital versions theKronecker delta is used.

or the identical expression

Using this definition an image containing two lines are transformedinto the Radon transform shown to the right

It can be seen that two very bright spots are found in the Radontransform, and the positions shown the parameters of the lines in theoriginal image. A simple thresholding algorithm could then be used to pick out the line parameters, and given that the transform is linearmany lines will just give rise to a a set of distinct point in theRadon domain. In my Ph.D. thesis the relationship with the Houghtransform is investigated, and it is shown that the Radon transformand the Hough transform are related but NOT the same.

Noise

The very strong property of the Radon transform is the ability to extractlines (curves in general) from very noise images as shownbelow. Theoretically results regarding the influence of noise can befound in Chapter 5 of my Ph.D. thesis.

In general many lines hidden in an image can be transformed into aset of peaks, where the value in the Radon domain (to the right)reflect the value on the individual lines. From the Radon transform,shown to the right, it can be seen that crossing lines makes noproblem.

The Generalized Radon transform

It is possible to generalizethe Radon- (and Hough) transform in order to detect parameterizedcurves with non-linear behaviour. In chapter 4 of my Ph.D. thesis afast algorithm can be found that used the generalized Hough transformto create irregular regions in the parameter domain corresponding tothe parameter regions of interest. Subsequently, it uses thegeneralized Radon transform within these regions in order to estimatethe curve parameter with high resolution.