Hi,
Today we updated the HDRComposition tool with a new functionality that adapts the composition images to produce a seamless result in specially difficult cases. The main parameter, Replace large scales, performs a large-scale adaptation between the images; this way, the shorter exposure is placed over the large structures of the longer exposure. This adaptation is specially useful to correct bloomings. Below you can look at the effect of this parameter:
Without large-scale adaptation:
With large-scale adaptation:
Two crops at original size:
This HDR composition is the result of three master images with an exposure of 300, 60 and 10 seconds. The images are proprietary to Leo Bette.
This kind of composition problems come from two sources:
- The limitation in the scaling precision imposed by the high noise level of the shorter exposures.
- The residual image gradients, specially if they are different in the short and long exposures.
This new parameter tries to adapt the composition at a local level to achieve a seamless composition in these difficult cases. We use the multiscale median transform to transfer the large scales from the long-exposure image to the short one. The Replace large scales parameter sets how to split the images between small and large-scale components. By setting it to zero we simply disable this technique. A higher-than-zero value sets the number of multiscale layers to be removed to generate the large-scale image components. Because it does not make any sense to have a small value in this parameter, the current parameter value equals to that value plus 4. This means that, for a value of 2, we'll remove the first 6 layers to generate the large-scale images.
To work properly, we should be able to completely remove the saturated areas in the large-scale images. Here you have the large-scale images generated with MultiscaleMedianTransform by removing 4, 5 and 6 layers (remember that the parameter will have always a value of n - 4):
The large-scale images should have no traces of the saturated areas to avoid any artifact. When working with your images, you can check how many layers you need to remove by loading the image with bloomings and generating a large-scale image by disabling the first n layers in MultiscaleMedianTransform. For this composition we removed 6 layers, so we set the Replace large scales parameter to 2. Please note that this is very important; leaving the saturated structures in the large-scale images will generate artifacts in the composition. These are the settings of HDRComposition for this image:
A common problem with this kind of composition is the noise in the shorter exposures. Please keep in mind that you won't be able to have a good composition if your short-exposure master has a very short total exposure time, since its noise will be much stronger than the long-exposure master.
Noise was a problem for this image as well, so a denoising process was needed over the blooming areas. I recommend to activate the Output composition masks check box to use the composition masks in the denoising process. For this image, the denoising consists of a MultiscaleMedianTransform process followed by a MorphologicalTransformation process; these are the settings of both tools:
Below you have a side comparison between the original composition and the composition after denoising:
This technique requires more exposure time since we need to acquire short-exposure subframes, but the resulting composition has no painting at all and recovers the true information behind the bloomings. For instance, in the image above, we reconstruct the large scale components of the dark nebulae in the composed image. We recover some faint stars as well, as seen below:
Aside from this new parameter, please note that as a general rule for blooming correction we'll need:
- To increase the Mask growth parameter to grow the composition mask, since the bloomings usually have non-saturated (or even dark) edges.
- We'll always need to smooth the composition mask since the composition never will be absolutely perfect.
Hope you'll enjoy this new feature.
Best regards,
Vicent.
Today we updated the HDRComposition tool with a new functionality that adapts the composition images to produce a seamless result in specially difficult cases. The main parameter, Replace large scales, performs a large-scale adaptation between the images; this way, the shorter exposure is placed over the large structures of the longer exposure. This adaptation is specially useful to correct bloomings. Below you can look at the effect of this parameter:
Without large-scale adaptation:
With large-scale adaptation:
Two crops at original size:
This HDR composition is the result of three master images with an exposure of 300, 60 and 10 seconds. The images are proprietary to Leo Bette.
This kind of composition problems come from two sources:
- The limitation in the scaling precision imposed by the high noise level of the shorter exposures.
- The residual image gradients, specially if they are different in the short and long exposures.
This new parameter tries to adapt the composition at a local level to achieve a seamless composition in these difficult cases. We use the multiscale median transform to transfer the large scales from the long-exposure image to the short one. The Replace large scales parameter sets how to split the images between small and large-scale components. By setting it to zero we simply disable this technique. A higher-than-zero value sets the number of multiscale layers to be removed to generate the large-scale image components. Because it does not make any sense to have a small value in this parameter, the current parameter value equals to that value plus 4. This means that, for a value of 2, we'll remove the first 6 layers to generate the large-scale images.
To work properly, we should be able to completely remove the saturated areas in the large-scale images. Here you have the large-scale images generated with MultiscaleMedianTransform by removing 4, 5 and 6 layers (remember that the parameter will have always a value of n - 4):
The large-scale images should have no traces of the saturated areas to avoid any artifact. When working with your images, you can check how many layers you need to remove by loading the image with bloomings and generating a large-scale image by disabling the first n layers in MultiscaleMedianTransform. For this composition we removed 6 layers, so we set the Replace large scales parameter to 2. Please note that this is very important; leaving the saturated structures in the large-scale images will generate artifacts in the composition. These are the settings of HDRComposition for this image:
A common problem with this kind of composition is the noise in the shorter exposures. Please keep in mind that you won't be able to have a good composition if your short-exposure master has a very short total exposure time, since its noise will be much stronger than the long-exposure master.
Noise was a problem for this image as well, so a denoising process was needed over the blooming areas. I recommend to activate the Output composition masks check box to use the composition masks in the denoising process. For this image, the denoising consists of a MultiscaleMedianTransform process followed by a MorphologicalTransformation process; these are the settings of both tools:
Below you have a side comparison between the original composition and the composition after denoising:
This technique requires more exposure time since we need to acquire short-exposure subframes, but the resulting composition has no painting at all and recovers the true information behind the bloomings. For instance, in the image above, we reconstruct the large scale components of the dark nebulae in the composed image. We recover some faint stars as well, as seen below:
Aside from this new parameter, please note that as a general rule for blooming correction we'll need:
- To increase the Mask growth parameter to grow the composition mask, since the bloomings usually have non-saturated (or even dark) edges.
- We'll always need to smooth the composition mask since the composition never will be absolutely perfect.
Hope you'll enjoy this new feature.
Best regards,
Vicent.
