Tuesday, 30 September 2014

Image Quality Tool - Modified Images - Before and After


Quality is all about consistent standards.  Normally, in terms of photographic quality, we mean an aesthetically pleasing image with good exposure and white balance.  But, when it comes to bird identification we are referring to something slightly different.  An image may be of an acceptable quality overall, but the subject of interest may be skulking in the shade, or bathed in the green glow of a foliage canopy.  More often than not, if an image of a bird presents an identification challenge it is because the overall image quality is poor - i.e. the image is not well exposed, the focus is off, and/or the white balance is wrong.  So, frequently bird identification images can do with some additional photofinishing to bring out salient identification features or rule out a similar species. 

Here I have taken a selection of very mediocre to poor record shots and used a recognised photofinishing tool (Adobe Elements 12) to try and improve the images for identification purposes.  I have particularly focused on digiscoped and video grab images in this posting.  

I am interested in the effect of these modifications on the scoring of the images in the Image Quality Tool, so I compare the original, unaltered JPEG image (left) with the best I could manage in Elements 12 (right).  I have included the scoring of both images and some detail around the images, how they were created and how they were modified.  

Note, in order to avoid too much lossy compression I started by saving the original JPEG as a PNG file, which is the lossless, compressed format I use for all files on this website.  The major advantage with PNG over JPEG is that changes can be made and the PNG image can re-saved repeatedly with minimal loss of quality.  Constant resaving of JPEG files results in progressively more compression, more artefacts, and more, permanent loss of data with each and every save.  If your images are JPEG it is really important to protect the original files by making working copies for modification and leaving the original file stored in a safe place.  Avoid at all costs modifying and re-saving the original JPEG files unless you are comfortable with the loss of fidelity and an increased level of JPEG lossy compression artefacts.

Proving that a bird may be identifiable even from the poorest quality image, this moulting adult White-winged black tern was captured as a video grab.  Aside from the low resolution and poor focus there are very obvious and disruptive artefacts including lossy compression and "combing" artefacts caused by poor deinterlacing of the video.  When making video grabs it is really important to use a good software package that deals with interlaced video properly.  There really is very little that can be done to improve the quality of this image, other than a slight white balance correction.

Older format video grabs tend to be of poor quality. Low resolution, sharpening halos, purple fringing and compression artefacts are the order of the day.  This grab was taken with a Sony DCR-PC330E which uses Mini DV tape format and Sony's bespoke media card, Memory Stick.  It has the facility to create video grabs on the go which are saved to the memory stick as compressed JPEGs. It does a pretty good job deinterlacing the video when creating grabs and this is about as good as we might expect from this, by now, well outdated, 1990's technology.

Swifts are notoriously difficult to photograph correctly.  Almost invariably, images of swifts photographed against the sky will be underexposed, unless great care is taken with Metering and Exposure.  Here I have increased brightness, contrast and clarity (mid-tone contrast) to try and bring out the contrasting throat/breast area.  Interestingly, both the modified and original image score the same using the tool.

High Dynamic Range lighting is very challenging for observation and photography.  While the overall or net image exposure of the original image was slightly underexposed, parts of the scene are moderately overexposed (in sun) and other parts drastically underexposed (in shade).  Digital cameras have a narrower dynamic range when compared with human vision, so photography on very bright days can be very hit and miss.  When scoring image quality we are looking only at the subject.  In this case we see major exposure issues (both over and underexposure) so this image scores a negative 10 points on exposure.  Although this is perhaps the only major fault with the image, it is a really big deal from an identification point of view, given the species involved, so the image requires a major overhaul in photofinishing terms to draw out salient features.  

Working with an original RAW image offers some hope of retrieving detail and colour from shadows and highlights.  In fact, the exposure of his particular image suits this process well (see ETTR HERE).  However the camera involved (Kyocera U4R) does not generate RAW files for edit and so we only have JPEGs to work with.  JPEG images have a much lower tolerance for major exposure adjustment and, consequently it took a lot of effort to draw out details and colour from this image in Adobe Elements 12.  The modified image looks pretty terrible and unnatural, though at least it allows for an analysis of some the finer ID points.

This image demonstrates the benefits of using photofinishing tools for ID from photographs.  But in this case, the image actually scores more poorly following modification, in particular because image noise, blooming and sharpening halos have become more pronounced in the final edit.  The Image Quality Tool only serves to measure image quality parameters - it takes no account of how identifiable the bird is in one image versus another.

A word of caution!  With such a high level of image manipulation there is a high level of risk and with that comes responsibility.  If I was to receive the right hand image in an email for ID consideration without any awareness that the original image was high in contrast and had been heavily manipulated, it could be very misleading.  Perhaps there should be a health warning on such images.  During analysis, it is always best to start with the original image and work with that so that you can understand the context in which the changes were made and their affects upon the image.  Occasionally, I come across unnatural-looking images online which must have undergone a lot of work, but it can be hard to judge what exactly was done or why.

Another nasty, underexposed image, owing to a combination of a dull day, a shy bird and digiscoping (this time the Nikon Coolpix 4500 with a Leica scope).  Standard modification for an underexposed image is to brighten it, increase contrast and perhaps adjust white balance.  In this case there is no net change in image quality, though identification clues are a small bit easier to see.

The "auto levels" function in Elements 12 did a very good job in correcting brightness, contrast and white balance in this image.  As a final adjustment I toned down saturation a bit and sharpened the image slightly using unsharp mask, to mask slight motion blur and improve image acutance (HERE).  The net effect is a significant improvement in image quality and perhaps a slight improvement in the display of key identification features.

Photographing birds under a canopy of foliage might seem like a complete waste of time, but HERE I was able to demonstrate that the diffuse yellow-green light transmitted by foliage can be surprisingly uniform in colour and can therefore be corrected for using a proper white balance tool.  Here I was able to use the white balance dropper with a point on the bird's white and grey flank feathering to reach a fairly neutral white balance position. 

The canopy is not completely closed.  Consequently there is some extremely bright dappling of the crown, mantle, tertials and tail.  Occasionally, I see dappled light and shade like this being referred to as a digital artefact and this is incorrect.  This is merely a lighting effect and is intrinsic to the original image - i.e. it is not an image distortion.  Such bright light can however result in blooming artefacts and there is probably some of that going on alright.  For more on this see HERE and HERE.

Though the modified image may not be perfect, it is much better than the original and deserving of the much improved image quality rating it receives.

2005 was a good year for Grey-cheeked Thrush in Ireland!  The original image is a pretty typical digiscoping JPEG.  I am not quite sure why many camcorder and some older compact digital cameras produced dull, greenish digiscoped image files.  The cast has all the hallmarks of a white-balancing error and can easily be resolved, but I suspect it may actually be a colour cast picked up from the scope or perhaps an indication of inferior demosaicing or deinterlacing interpolation by the camera processor (more HERE and HERE).   I hope to establish it's causation at some point as it certainly annoyed the hell out of me for years! 

Once again "auto levels" in Elements 12 has done a great job with this image and I topped it off by reducing colour saturation slightly. There is a slight net improvement in image quality and the bird is certainly more clearly identifiable as a result.

No comments:

Post a Comment