Talk:Thresholding (image processing)

Source of given algorithm?

Isn´t this the "ISODATA" algorithm by Ridler & Calvard?

Ridler TW, Calvard S. (1978) Picture thresholding using an iterative selection method, IEEE Trans. System, Man and Cybernetics, SMC-8: 630-632.

Credit should be given to the original source! Maybe it´s an indepent "re"-invention?

???

I believe the improved threshold should be: (||G2||*m1+||G1||*m2)/(||G2||+||G1||).

Last paragraph of multiband thresholding

The paragraph of the section in multiband thresholding:

In histology images the colour is given by 2 or more dyes, for example hematoxylin and DAB. The contributions of these two dyes to the colour image can be established by using colour deconvolution. The incident light is measured by looking at an unstained part of the slide. Then you measure the relative absorption of each the 3 RGB components of the image by each dye. Absorption is defined by Beer's Law. The absorption characteristics for each dye can be obtained by measuring the RGB components of the colour in areas of the slide that are only coloured by one dye. The computer software can solve 3 linear equations for each pixel in the image, one for each colour channel, in order to resolve the amounts of each of the dyes at each locus in the image. This method can resolve up to 3 different dyes. The result is an image for each dye, showing the amount of that dye throughout the image. This allows the amount of dye to be quantified, which is important for image analysis of immunohistochemical stains.

I'm unsure what this has to do with multi-band thresholding. it may be an example of multi-band thresholding, but there is too much background information; should this be it's own stub? Albert Cruz 19:51, 1 September 2010 (UTC)

Weighted average doesn't make sense

I removed the following:

Another approach is to calculate the new threshold in step four using the weighted average of $m_{1}$ and $m_{2}$: T’ = (||$G_{1}$||*$m_{1}$ + ||$G_{2}$||*$m_{2}$)/(||$G_{1}$||+||$G_{2}$||), where ||$G_{n}$|| is the number of pixels in $G_{n}$. This approach often gives a more accurate result.

because it appears to do nothing more than calculate the mean value of all the pixels, independent of the threshold value. The mean of $G_{1}$ pixels times the number of $G_{1}$ pixels is the sum of $G_{1}$ pixels. Same goes for $G_{2}$. The union of $G_{1}$ and $G_{2}$ is the whole image, so the sum of those products is the sum of all the pixels in the image and ||$G_{1}$|| + ||$G_{2}$ is the number of pixels in the image. Ergo, their quotient is the mean of all the pixels in the image.

Bryan Henderson 22:15, 20 May 2007 (UTC)