Frequency Separation

[Round Tuit]

The recent debate in the thread Lady in Red on the merits of frequency separation led me to investigate the procedure. I am not a portrait photographer and I mostly avoid taking candid shots of people. I can see though that the procedure could be very helpful in retouching flower petals and other subjects where a fine texture is superimposed on a coarser tonal structure.

My understanding, so far as it applies to Photoshop, is that the image is duplicated twice. A Gaussian blur is applied to the bottom layer which now contains only the low frequency components of the picture. That low frequency layer is subtracted from the upper layer which becomes the high frequency layer and its blending mode is set to "Linear Light". The result is that the stacks of these two layers is an exact duplicate of the original image.

Retouching is done on the individual layers. Colour and tonal variations can be corrected on the low frequency layer without affecting the fine details in the image. Similarly, the texture can be retouched on the high frequency layer without affecting the colours. As far as I know, there is no restriction on the techniques used to do the retouching of the layers. Cloning, healing, dodging and burning on the layer as well as adding intermediate layers, with or without mask, can all be used to retouch each layer.

The relationship between the two layers is fixed by the radius of the Gaussian blur used. An extremely large radius will put the entire image in the high frequency layer. A radius of zero will put the image in the low frequency layer. In either case you gain nothing. A radius that only blurs the fine texture will give the greatest benefit.

I conclude from the above that the procedure can make it much easier to retouch textured images and at worst provide no benefit at all but it cannot by itself damage the image. So, what am I missing because this conclusion is different that the one reached in the "Lady in Red" thread?

[xpatUSA]

The recent debate in the thread Lady in Red on the merits of frequency separation led me to investigate the procedure. I am not a portrait photographer and I mostly avoid taking candid shots of people. I can see though that the procedure could be very helpful in retouching flower petals and other subjects where a fine texture is superimposed on a coarser tonal structure.

My understanding, so far as it applies to Photoshop, is that the image is duplicated twice. A Gaussian blur is applied to the bottom layer which now contains only the low frequency components of the picture. That low frequency layer is subtracted from the upper layer which becomes the high frequency layer and its blending mode is set to "Linear Light". The result is that the stacks of these two layers is an exact duplicate of the original image.

Retouching is done on the individual layers. Colour and tonal variations can be corrected on the low frequency layer without affecting the fine details in the image. Similarly, the texture can be retouched on the high frequency layer without affecting the colours. As far as I know, there is no restriction on the techniques used to do the retouching of the layers. Cloning, healing, dodging and burning on the layer as well as adding intermediate layers, with or without mask, can all be used to retouch each layer.

The relationship between the two layers is fixed by the radius of the Gaussian blur used. An extremely large radius will put the entire image in the high frequency layer. A radius of zero will put the image in the low frequency layer. In either case you gain nothing. A radius that only blurs the fine texture will give the greatest benefit.

I conclude from the above that the procedure can make it much easier to retouch textured images and at worst provide no benefit at all but it cannot by itself damage the image. So, what am I missing because this conclusion is different that the one reached in the "Lady in Red" thread?

André, I have not used Photoshop, so I have no detailed comments to offer other than the opinion that you have missed nothing.

I myself use similar processing in either the GIMP or RawTherapee where it is called "wavelet processing" and has fixed levels of 1px,2px,4px,8px etcetera up to I think 1024px. The GIMP is a little easier because it splits the detail levels (default 5) into the number of levels as layers plus a "residual" layer - each layer editable by any means desired.