I was looking at the comparison of 2007 and 2015 HST images of the "disappearing" star N6946-BH1 in Phys.org's article Collapsing star gives birth to a black hole. The 2007 images is from the WFPC2 while the one from 2015 is from WFC3.

I wondered why the WFPC2 image is substantially fuzzier than that from WFC3. I looked at the WFPC2 manual for cycle 17 and found it has the "L-shaped" detector. Three quadrants have CCD elements with pixel spacing of 0.1 arcsec, the fourth "planetary" array has a pixel spacing of 0.046. However it seems more complicated because - if I understand correctly - there are different focal ratios used for the two.

The WFPC2 field-of-view is divided into four cameras by a four-faceted pyramid mirror near the HST focal plane. Each of the four cameras contains an 800x800 pixel Loral CCD detector. Three cameras operate at an image scale of 0.1″ per pixel (F/12.9) and comprise the Wide Field Camera (WFC) with an “L” shaped field-of-view. The fourth camera operates at 0.046″ per pixel (F/28.3) and is referred to as the Planetary Camera (PC).

However, while the WFC3 manual for cycle 23 gives the pixel spacing as

Table 5.1: WFC3 Detector Characteristics

0.04″ per pixel

I didn't find a focal ratio.

I'd like to understand why the 2007 image is so "gaussian fuzzy". If it were only because of larger pixel spacing it would look more blockish or pixellated. Does this mean it was using a different optical path with lower optical resolution? Is the 2007 image from the WFC part at F/12.9? Does the WFC3 use F/28.3?

I also don't understand the optical reason why different paths would have different optical resolutions. This would not come from different Airy disks from simple diffraction - the angular size would be unchanged.

enter image description here enter image description here

above: cropped sections from the same image shown below.

enter image description here

above: From here.

  • 1
    $\begingroup$ Have you established whether the image comes from the WFC CCDS? If so, the angular pixel scale is 2.5 times worse than that of WFC3. Wouldn't that explain it, because 0.1 arcsec/pixel does not sample the PSF sufficiently? The images are not "blocky" because they are "drizzled". $\endgroup$
    – ProfRob
    Jun 4, 2017 at 12:30
  • $\begingroup$ @RobJeffries no I haven't and I'm not sure how to track that down. I don't even have a scale for the image. I understand what your mean though, and that could explain it all. Mathematically, if "drizzling" is a special kind of convolution/filter that "de-blockifies" the image to round de-focus, that could be really interesting mathematically just by itself. I'll keep poking around and see what else I can find, an image processing answer does sound much more likely than an optical answer. Thanks! edit: search for drizzingling going well! stsci.edu/hst/wfpc2/analysis/drizzle.html $\endgroup$
    – uhoh
    Jun 4, 2017 at 12:49
  • $\begingroup$ A quick check with the STScI HST archive shows that the pre-SN images did indeed have the progenitor star on one of the WF chips of WFPC2, so Rob's suggestion is correct. If you look at Figure 1 of the paper (arxiv.org/pdf/1609.01283.pdf), you can see that the pixels in the WFPC2 images are larger than the WFC3 images. $\endgroup$ Jun 21, 2017 at 9:41
  • $\begingroup$ @PeterErwin I did a 2DFFT on a chunk of the WFPC2 image above to see if I could discover some residual evidence of the drizzling algorithm, just for fun and recreation, but soon realized what I saw could mean all kinds of things. Figure 1 on the other hand is really conclusive! Would you consider posting your comment as an answer? It certainly looks conclusive. Possibly with a screen shot? i.stack.imgur.com/zdZi3.png $\endgroup$
    – uhoh
    Jun 21, 2017 at 9:50
  • 1
    $\begingroup$ @uhoh -- Sure; give me a few hours... (Yeah, the color images above are the combination of 2 or more single-filter images, with possibly some extra processing to make the colors looks nice, so it might be hard to get back to the original images that way.) $\endgroup$ Jun 21, 2017 at 10:18

1 Answer 1


As Rob Jeffries suggested, in the WFPC2 observations the progenitor star fell located on one of the wide-field ("WF") chips of WFPC2. (I verified this by downloading one of the preview images from the Mukulski Archive.) This was purely by chance, since the WFPC2 observations were taken years before the discovery of the failed supernova. It's good luck that the region of the galaxy where the failed SN took place was imaged by WFPC2, and only slightly bad luck that it wasn't in a region imaged by the planetary-camera ("PC") chip.

WFPC2 is/was an array of four equal-sized CCD chips, each with 800 x 800 15-micron-wide pixels. However, the optics were set up so that 3 of the chips (the WF chips) were imaged using a focal ratio of F/12.9, while the fourth (the PC chip) was imaged separately with a focal ratio of F/28.3. In essence, a magnified optical beam was focused on the PC chip, giving a scale of about 3.0 milli-arcseconds per micron, or 0.0455 arc seconds per pixel. Meanwhile, the WF chips received a scale of 6.67 milli-arcseconds per microns, or 0.097 arc seconds per pixel. This undersamples the actual resolution of the telescope (which has a point-spread function with a full-width half-maximum of about 0.09 arc seconds in the I-band).

The UVIS (UV-visual) part of WFC3 has two 2k x 4k CCDs (with, I think, 15 x 15 micron pixels) and a focal ratio of F/31, which works out to 0.0395 arc seconds per pixel.

So in practice, WFC3 will always resolve things better than the WF chips of WFPC2 can, because its pixels can properly sample the telescope resolution and the WF pixels cannot.

You can see the difference in the upper "progenitor" (WFPC2) images, compared with the lower "2015" (WFC3/UVIS) images from the actual paper:

From Figure 1 of the original paper

Note the stronger visible pixelation in the WFPC2 images (top two panels), due to the fact that the WF pixels are larger in angular size.

I suspect any other apparent differences in the press-release images are probably due to the processing used to create the color image. In particular, the side-by-side (WFPC2 vs WFC3/UVIS = "before and after") color image from the press release shows both images at the same spatial scale, as part of a single image file, which means that the same pixel density is present throughout the combined image. But since each original WF pixel is about 2.5 times the (angular) size of a WFC3 pixel, the WFPC2 image had to be rebinned to match WFC3 pixel scale (1 WF pixel = 6.25 WFC3 pixels) before the combined (JPEG/PNG/TIFF) color image could be made. The interpolation used for this rebinning process may be what introduced the extra smoothing in the WFPC2 image.

[Edited to add comment about rebinning used to generate color press-release image.]

  • $\begingroup$ Thanks for the post! @RobJeffries had also mentioned drizzling (dithering, convolution, blurring) in response to my comment I'd like to understand why the 2007 image is so "gaussian fuzzy" There's even a page for it. Maybe include at least a cursory mention of drizzling? I'm guessing it may have been a fairly standard thing to do to images used for public release, rather than used as data for analysis. A follow-up question might address the details of drizzling later. stsci.edu/hst/wfpc2/analysis/drizzle.html $\endgroup$
    – uhoh
    Jun 21, 2017 at 21:13
  • 1
    $\begingroup$ The grayscale images in the paper (part of the data analysis) were produced via drizzling, too, so I'm not sure that's the whole explanation for the apparent fuzziness of the press-release images. It's possible the production of the latter involved some additional smoothing to suppress the noise and make them look nicer.... I'll see about adding a brief mention of drizzling to the answer. $\endgroup$ Jun 22, 2017 at 12:26
  • $\begingroup$ OK great. I'll look into it too. The images in the question are very fuzzy, the ones in the answer show clear pixelation. The whole reason I asked the question in the first place is that there was no visually detectible evidence of pixelation, the images look round and "gaussian-fuzzy". The central motivation for the question is the "any other differences in the press-release images". The answer may in fact not be drizzling proper; in this case there may be a separate post-processing routine for public release photos! $\endgroup$
    – uhoh
    Jun 22, 2017 at 15:01

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .