One of the issues I (and, I expect, others, based on what I see) have in downsizing for the BLP website is controlling haloes in resized images. This effect seems to vary depending on the particular features of the image, being most prevalent around hard high contrast edges such as beaks and feet against neutral backgrounds. It also seems to be impacted by the amount of downsizing eg an image which is heavily cropped (and therefore downsizing to 1800 px does not involve too much adjustment) tends to show a weaker halo effect, compared to one which is downsized from an uncropped image. In more severely affected images I can sometimes minimise it by reducing sharpening in the original image before downsizing.

From what I read online this is a common and perhaps even inevitable effect of the downsizing process. The software I use defaults to bicubic algorithms for downsizing, either just plain bicubic or bicubic-sharper, which still seems to be the basic system used by many software packages. The lanczos algorithm seems to have similar effects. Beyond this there may be specialised proprietary methods which are 'smarter'. As far as I can tell from a quick examination, Lightroom's downsizing doesn't seem to specify how it does downsizing, althought it offers several options in terms of the amount of sharpening applied. I haven't noticed a huge difference between selecting bicubic or bicubic sharper, though perhaps the halo effect is more pronounced in the latter in some images.

Perhaps members can share practical tips on how they minimise downsizing halo artifacts.

Simon

Hi Simon

Can you outline what software you are finishing and downsizing in when you see this? In Lightroom at least, sharpening should be applied as part of the final image save when downsizing occurs but as far as I know, the user has no control over that sharpening. I've never seen a halo like this when downsizing in Photoshop Elements (where I can apply and control sharpening after downsizing occurs). What you are seeing sounds a lot like the sharpening halo seen from relatively strong use of unsharp mask which I never use after downsizing (instead use the Sharpen: Remove Gaussian blur option).

I don't believe I see this issue in my work. Maybe I need to look closer.

My output workflow is:

(i) downsize the hi-res version of the image (a 16- or 8-bit TIFF) to the finished size using bi-cubic interpolation. The image will be a little 'soft' due to the blur introduced by the downsize algorithm. The blur is similar to Gaussian blur. There should not be any halos showing at this stage.

(ii) select all the bird using a selection saved earlier in the workflow or select part of the bird requiring output sharpening. Remove the blur inside the selection using 'remove Gaussian Blur' deconvolution found in PS. I find amount = 50, radius 0.3 - 0.5 will remove the downsize blur and restore the sharpness.

(iii) the image will still be a 16- or 8-bit TIFF at this stage and needs to be converted and saved as a JPEG with minimal compression and maximum quality.

Any downsize option that includes global output sharpening by default should be avoided.

Thanks Ian and Glenn

I agree that your photos don't show this effect, which might be because of your workflow. I am seeing it in some other photos on the BLP site for example Diana Womersley's most recent Moderators Pick (Jacky Winter) shows distinct halo effects around the bill and legs, which might be due to sharpening during raw processing but might also be due to downsizing (or both) (apologies Diana for picking on you, its just a convenient recent example). In my own photos it occurs in some situations when downsizing, noticably around very abrupt transitions like bills, and (as I said) can be reduced in various ways. I just have to keep an eye on it.

I have found a number of sites which refer to an 'overshoot' or 'ringing' problem (which I understand are not quite the same thing) from resampling algorithms. For example, this from Wikipedia's article on acutance: "Low-pass filtering and resampling often cause overshoot, which increases acutance, but can also reduce absolute gradient, which reduces acutance. Filtering and resampling can also cause clipping and ringing artifacts." A Cambridge in Colour article on image resampling states: "All non-adaptive interpolators i]SP - this includes bicubic[/i attempt to find an optimal balance between three undesirable artifacts: edge halos, blurring and aliasing. Even the most advanced non-adaptive interpolators always have to increase or decrease one of the above artifacts at the expense of the other two — therefore at least one will be visible." Cambridge in Colour includes a separate tutorial on reducing moire when downsizing for the web. I found some other articles a while ago but can't bring them up at the moment.

Perhaps your technique involves a less sharp starting point, and results in more blurring during the transition; this would reduce halos and moire, but places more emphasis on post-resampling sharpening?

Glenn, in answer to your question I have used two techniques and the effect can be seen in both. First, I use DxO Photolab, which includes a bicubic or bicubic - sharper option. I am tending to use this now that we have a square 1800px maximum allowance; DxO makes it awkward to downsize to a maximum pixel size specified in both the horizontal and vertical dimensions, as it only allows you to specify a dimension for the longest edge. In terms of sharpening during raw processing ie before downsizing, DxO's main tool is their proprietary lens/camera measurements/corrections which automatically correct for lens issues. DxO has an unsharp mask tool which is off by default and I never use it.

Second (when we specified maximum dimensions in both X and Y axes) I found it easier to use Lightroom for downsizing. This involves exporting a full size 8 bit TIFF from DxO to Lightroom (this happens directly via a plugin) and then downsizing and exporting a JPEG from Lightroom. Although this seems clumsy, once set up it literally takes just a moment. Adjustments to sharpening can be made to the TIFF in LR if one wants. You are correct in that Lightroom (at least the version I have) does not give you much specific control over sharpening in the downsizing process, although it has a few generic settings.

Simon

Hi Simon,

I think the source of the problem with Dianna's Jacky Winter is too much global sharpening in her workflow. If she used the approach described in my Notes using optimal global sharpening during capture sharpening (not USM) and a selection at the end for output sharpening using the PS 'reduce Gaussian blur' option, then there would be no halos on hard edges.

As far as I can see, downsizing using bi-cubic resampling per se should not exaggerate existing edge halos but global output sharpening surely will.

Regarding your workflow, I don't understand why you choose to do the finishing in LR when PS gives so much more flexibility. I think DxO Photolab is excellent for RAW adjustment and conversion. It has top class NR and capture sharpening based on measured lens + camera deconvolution kernels, in principle as good as Canon DPP. I would take advantage of these capabilities and export a 16-bit TIFF to PS for finishing and cataloguing rather than to LR.

That's interesting. For Diana's shot of the Jacky Winter, the bill halo is quite subtle on my screen - in fact, I never noticed it until you pointed it out and then I had to look with a magnifier at my screen. It was even harder to spot the one on the front of the leg and nothing visible at the back of the leg. I assume you are viewing these all at 100% on your screen Simon? That is one pixel in the image = one on the screen? Anything else will make it difficult to tell what is going on.

As for the downsizing and global sharpening in Lightroom, does the problem look more obvious than downsizing and output sharpening in DxO? The only other thing I can suggest is using Canon Digital Photo Professional to process the RAW and compare the result with what you are getting from DxO. For DPP, set picture style to Neutral, Sharpness to 2.5, USM off and Digital Lens Optimiser to 50%. They are the settings I use, as recommended by Ian who has done the hard work to test and set these up. Using DPP might give some hints as to the source of the issue.

As for the specific Bicubic algorithm for downsizing, there are three options in Photoshop Elements. I use Bicubic (best for smooth gradients). The other options, which I don't use and haven't tried are Bicubic Smoother (best for enlargements) and Bicubic Sharper (best for reduction).