the unsharp mask Let's first understand how the most important sharpening tool, the unsharp mask works. The diagram shows how an image has a sharp boundary from light to dark grey as shown in the solid blue curve, but the photograph shows a soft boundary as in the red solid curve. If we could apply a correction as in the blue dashed curve, it would be fully sharp again, but such a correction cannot be made from thin air. The unsharp mask blurs the image further as shown in the green solid curve, but this intermediate stage is not visible. It only exists in a patch of pixels given by the radius parameter. Think of it as a sliding window. For each pixel the algorithm now calculates the difference with the original curve (in black) and adds a percentage of this difference to the image (red dashes). As can be seen, the image appears to have sharpened but at the cost of an unwanted side effect, a kind of bump or halo as the dashed red line overshoots. Note also that these two bumps increase contrast which makes the image appear sharper.

 
Note that film grain comes in many sizes. It so happens that large silver crystals are more sensitive to light. Thus the largest grains are found in the darkest tones. When shooting with neg film it therefore pays to overexpose your film as a matter of principle, by 0.5-1.0 f-stops. This also brings better colour rendition into the dark shadows, but it also brings more scanner noise in the highlights and it may affect sharpness by a slight halo effect. This is the reason that some scanners do a very poor job of neg film, something that does not show in their official specifications. The best scanners do a pre-scan which adapts scanning speed such that dark images take longer to scan than bright ones, and signal quality is not compromised.
It is also a fact that CCDs produce the most noise in the darkest signals, which correspond with the dark areas in slide film and with the highlights in neg film. Thus neg film scans suffer from noise both in their highlights and shadows. Fortunately the human eye is most sensitive to noise in the mid tones.

Note also that both the colour film and CCDs have a problem with red colours which are less energetic than the greens and blues. The red wavelengths have less quantum energy to free electrons from either the emulsion crystals or the CCD pixels. As a result their digital signals become more prone to noise. In underwater photography with negative film, this may pose a problem.

• Open the image you want to smart-sharpen
• do not convert to grey scale yet, as the contours alogrithm will find colour contours as well
• save as mask.jpg in the same resolution as the original. We save it here now so you won't accidentally destroy your original.
• effects/find-countours/edge detect. Adjust the parameters for best edge detection effect.
• auto-adjust for maximum contrast and convert to grey scale
• use the S-curve to exaggerate contrast and to exclude minor edges and to create blacks and whites.
• blur the image with a slight 1-2 pixel Gaussian blur or 3-4 pixel jaggy despeckle.
• save it (as mask.jpg)
• open your original image
• load mask from file: mask.jpg
• invert mask if necessary such that the contours are white for sharpening
• apply unsharp mask in the normal way. Judge the result - magic!
• if necessary, invert the mask and apply a low pass blur or jaggy despeckle to the smooth parts of the image
• save as a new photo. Don't overwrite your original!

f032809: photo of a giant heart urchin in uneven lighting. The original shows its belly spines almost over exposed, such that contrast could not be enhanced further. A gradient mask was made and the light part darkened by 20 units. Then contrast and lightness were enhanced to give the balanced result on right.

f032922: a male demoiselle guarding his nest with eggs. In the original photo (left) the foreground was too bright compared to the background and the colour of the fish. A gradient mask was drawn and the lower half darkened by 15 units. After inverting the mask, the top half was brightened by the same amount. A small S-curve effect was applied with the tone tool. The corrected image (right) has more colour, is better balanced and shows more detail in the fish. The whole operation took no more than 2 minutes to do.

The most important option of the mask is feathering. It is a kind of blurring which can extend across hundreds of pixels. It can be directed across the middle of the mask or inward or outward. It enables us to apply the effect gradually.
The next most important option is paint on mask. This treats the mask as a standard grey scale image on which you can paint, rub out, blur and do anything else. By turning paint on mask off again, the photo editing program works on your colour image again.

Area masking can be used for bringing the bright sky in line with the darker field. Take the freehand mask making tool and draw a mask around the sky on you image. Irregularity is better than a straight line. Then select feather and blur the mask inward by the amount required (say 10 pixels). Note that these 10 pixels produce ten steps from 0 to 255. You can blur these steps further by selecting paint on mask, then effect/blur/Gaussian blur and blurring these steps till they look smooth. Deselect paint on mask again. The mask overlay shows you what the mask is doing. Now apply the tone curve tool and darken the sky by pulling the curve down in the middle. Achieve no more than 0.5 fstop.
Invert the mask so it works on the rest of the picture. Now use the tone curve tool to pull intensity up by 0.5 f-stop, until the desired effect is reached. It is natural to overdo this in the beginning. Save the mask and turn it off to study the artefacts in the transition region.

Since the mask overlays the sky, it can also be used to smoothen the sky while also sharpening the field. Retrieve the mask from where you saved it. Apply Gaussian blur or jaggy despeckle to the sky but always leave some noise for improved colour defintion.
Now invert the mask and sharpen the field using the unsharp mask. Check the boundary area for anomalies.
 
 
 

f021512b: the image as scanned with a linear curve and no further corrections.

f021512a: scanned with a z-curve; area mask applied and corrected for excessive reds.

Area masks can be applied in a subtle manner. Here the contour of the Snell's circle and the two 'holes' in it, was traced by hand, then feathered by 10 pixels for a smooth transition. The sky was darkened with the tone curve tool. The auto adjust could be used because it nearly got it right. However, because there are no whites in the image, it exaggerated the reds which also make the picture murky and dark. After colour correction the image was blurred with the jaggy despeckle since it was taken with high speed film. The image was then resampled to about 3000x2000 pixels and sharpened twice. In a good image often the small differences matter.

density-dependent masks Masks that are created from the intensity of the pixels have a flavour of their own, offering a wide range of possibilities. In the diagram we have drawn a density-dependent mask suppressing the bright mid tones where our eye perceives sharpness most. In the example the mask produces a grey scale image with exaggerated contrast (steep slopes in the tone curve). For the result, see the 'mid-tones' greyscale image of the cave diver above. Check whether the areas where sharpness is required show dark. Then use the mask to blur highlights and darks. Invert the mask to sharpen the mid tones. A similar density-dependent mask can be used to bring the sky in balance with the field, as described above.

Note! although it is possible to enhance poor images considerably, nothing works as well as starting with a good image!

It is important to remember that the resampling process is akin to averaging and blurring, resulting in an unsharp image which needs to be resharpened. Also remember that small images need to be oversharpened to appear naturally sharp to the human eye. Use the adaptive unsharp tool for this, with a 25% effect. Use it twice or three times to achieve sufficient sharpness.
 
 

greyscale By converting a colour image to greyscale, the size in memory is reduced threefold because each pixel represented by three bytes for RGB is now represented by a single byte. However, when saving as a JPEG compressed file, greyscale images are not remarkably more compact than colour images!! Thus for saving file space, it is not worth losing the colour of your image.

 

GIF paletted The GIF (Graphic Internet File) is a very clever standard for paletted images consisting of only few colours (255 max). The large 400x270 pixel diagrams on this page typically compress to less than 10KB whereas these would take 50KB in JPEG. However, for colour photographs the paletted form is totally unsuitable.

Superb compaction in GIF format is achieved by using few colours. Each pixel in a GIF file takes 1 bit for a two-colour image (say black and white but any two colours achieve the same), 2 bits for 4 colours, 3 bits for 8 colours and so on. Major compaction can be achieved by staying below 32 colours or half byte (= 4 bits) per pixel. The maximum number of colours is 255 but these can consist of any combination or palette (adaptive palette). Excellent compaction of line drawings like cartoons is achieved with 8 grey scales in 3 bits per pixel.

Where a drawing is done in few colours, the potentially compact file can become large if anti-aliasing is chosen which blends edges such that the alias or steps don't show. In our web site we have chosen to show alias (see the slanting lines in the diagram above) to achieve compactness.

Note! don't use GIF for photographs.

JPEG compression The JPEG standard (Joint Picture Expert Group) is used worldwide to compress colour RGB images or greyscale images. Allthough used by nearly everyone, I think that hardly anyone understands what it does and how to use it properly. JPEG compresses colour or greyscale images by making use of the way we see: luminosity (brightness) is more important than colour detail, and many pixels in the image look alike. The RGB colour profile is first converted to a YCC profile using a simple and fast algorithm (luminosity, colour 1, colour2) such that luminosity gains most importance (Y:C:C= 4:2:2).  Then it divides the images in 8x8 pixel cells, traversing each cell diagonally from top left to bottom right, looking for similarities. This 'Huff coding' compacts the bit stream in a reproducible way but it creates dissimilarities between cells, such that the cells become visible. Depending on the amount of compression desired, more or less detail is lost. Thus JPEG is considered a 'lossy' compression technique, which is true when driven to extremes. However, the original data file can be compressed 5-10 times without visible loss. The file can then be retrieved and saved using the same amount of compression, without it deteriorating further. So the combination of an sRGB profile with JPEG compression to 5-10x, is the way to go.

Example of extreme JPEG compression, where the 8x8 pixel square cells are clearly visible and the discontinuity created between adjacent squares. For more info: www.jpeg.org

So, JPEG does not need to suffer losses in picture quality if only little compression is used. One of the problems we have is that JPEG allows for a continuously adjustable rate of compression from 0 to 255, but what these levels mean is not immediately clear.

What we must realise is that JPEG compression does not lead to a predictable file size. If you have a busy image, a larger file results. If you have much pixel noise, the image becomes larger too. Ironically, removing noise from an image results in a smaller JPEG file! Compression does not produce the same losses on all files. Large files can be compressed more without noticeable loss, whereas small files cannot.

So what happens as one compresses an image further and further?
 
 
 

One could fear that losses may accumulate each time the image is altered and saved again. However, for compression rates of 10 and 20, this is not perceptible. Besides, the compression algorithm makes the same decisions as long as the image was not changed in size.

How do I recognise JPEG damage?

• The first signs of damage consist of a grid pattern appearing in your image (compression rate above 20). It is subtle but clearly visible at magnifications of 1:1 or higher. Ironically, it does not show as clearly in print, so for printers it should be of little concern. If needed, it can be removed by the jaggy despeckle blur. Similar damage sometimes happens as the result of resampling.
• The second sign is banding in colour gradients as the JPEG algorithm codes for larger colour steps with fewer bits (compression rate 40-80).
• The third sign consists of messy halos where density changes suddenly from dark to light (compression rate above 80). These are clearly visible and cannot be removed easily.

What is the JPEG header?
When you compact the photos from your scanner or digital camera, you may notice that they are reluctant to become more compact than 45-60KB, no matter what you do. This is caused by the fixed JPEG/JFIF metadata header placed there by the camera or scanner. This header contains all the photographic or scanner details (JFIF standard) and a small bitmap thumbnail picture of about 30KB, which is seldom used but confuses the display thumbnails option of Windows Explore. You want to get rid of this fast, but how?
You won't find a suitable option in Photoshop or Photopaint or other program, so you need a drastic solution in the form of the free JHEAD.EXE program from http://www.sentex.net/~mwandel/jhead/. Unfortunately it runs in command mode but it is very good at that and it is very small and does not interfere with Windows and it is available for Linux, Mac and other operating systems. Find it on the web and download it in the c:\windows directory. Windows will find it. It comes with a manual.

Suppose your compacted files are in the directory c:\mypix\  Proceed as follows:

• Start/run open the run dialog by clicking the Start button, then Run.
• Enter the following: jhead -purejpg c:\mypix\*.jpg
• It very quickly removes the JPEG headers from all JPG images in that folder. Now they are suitable for Internet. Thank you, Matthias Wandel.

Suppose you end up with file sizes averaging 3 or 6MB. Suppose you are able to back your files up on CD with a CD-ROM burner. Each CD reliably stores 640MB (don't be enamoured by 700MB as this limits playback on some systems). Thus your CD stores between 100 and 200 images. If you have a scanned library of 3000 images, this would require 10-20GB or 20-40 CDs. You may have this much space on your hard drive but when your scanned library reaches 15,000 shots, you will have most of your images 'off-line' on CDs. By that time it is handy to have a local library of comping images (900x600 pixels at compression rate 40 = 4-8% of your off-line library). So you might as well build this one up from the very beginning. They also give you a good slide show.
 
 
 

Recent storage technology has produced disk drives with capacities over 300GB! It may pay to acquire such a drive if you expect to produce many images. It can be housed in an external box connected to the PC by USB (Universal Serial Bus) or by firewire, and in this manner it also becomes portable and can be saved in a fire proof safe. This allows you to turn it off most of the time, which lessens its chance of self-destruction or corruption by Windows. But remember that a hard drive is no excuse for not making backups to CD.

Recent versions of windows allow you to view the contents of directories as thumbnails and this is very handy. However, the program (Explore) is rather primitive and cannot remember thumbnails very well even though it constructs massive thumbs.db files. To say it nicely, it is a time waster. Microsoft never appears serious as to the quality of its products. The Explore program, after over a decade of improvements, is still unable to print the contents of directories and likewise, the thumbnails. So you cannot use it to print a directory of thumbnails on paper, for inclusion in the CD jewel box or for making a paper reference to your stored images. Remember that scanning a paper index of thumbnails is very fast and you can scribble notes on it.
 
 
 

It is a pity that the JPEG standard does not allow for inclusion of keywords and a picture description. In fact, it does. But the standard is not adhered to and Photoshop eliminates the text. However, one day such anomalies will be ironed out and you will be able to include searchable text with every photo. The program to search and manage this facility already exists (Ujeta from www.moon1000.com) and if your picture editor retains JPEG comments, this is a way to go.

The original has been degraded by blurring. Note that the file size reduced from 70 to 50KB by being blurred. Make a mask for smart sharpening.

Signal noise, scratches and stains have been added to the blurred image and also an edge problem. The signal noise increased file size to 166KB. Remove all blemishes and try to blur and sharpen the image as well as you can. Note that the noise is uniform, thus not like film grain which is not as pronounced in the mid tones.

Colour and contrast have been changed. You must study the histogram to correct contrast and brightness. Colour correction is different for the highlights, mid tones and shadows. You may also need to resaturate. 65KB.

Density-dependent mask which can be used to sharpen and blur the mid tones in a different way, without losing sharpness in the contours. (58KB) See if you can make a similar mask.