Simple usage instructions
Here are described in a simple way the steps you need to do in order to filter an image with BlueNile.
To run the software double-click on the bluenile.bat file in the directory bluenile you created when installing BlueNile. Then push the 'Load' button to load an image you want to filter. Once the input image, it's frequency domain transform and the filtered image are displayed and the label 'Filter shape' is red, you are ready to start filter the image.
Tip: You can use the image kb108.jpg provided in the ..\bluenile\batch\ directory as a example image. It is electronradiography of the structure of a paper produced in the Low Countries around 1450-1500AD. The figure in the centre is a watermark, the horizontal lines are the so-called 'laid lines' - that is wires of the mould screen on which the paper pulp is deposited during paper manufacturing - and the two vertical lines are the 'chain lines' (other wires supporting the laid lines). More such images you can find on the web site of the Dutch National Library, from where our image originates.
The middle image ('FFT Magnitude') is a representation of the frequencies characterising the input image. It is called the Fast Fourier transform (FFT) and is composed of two parts - the magnitude and the phase - from which only the first one is shown by BlueNile. You can see, if you loaded the kb108.jpg file, two bright spots positioned on a vertical line on each side of the centre of the image. These two spots correspond to the number of horizontal lines on the input image (the laid lines): there are 49 laid lines spanning from one side of the image to the other and the brightest pixel of each spot is at 49 pixels far from the centre of the FFT image.
1. Suppressing frequencies
Suppose now that you are interested only in the watermark and you would like to have an image free laid lines. Point your mouse some 20 pixels NW of the centre of one of the two spots, hold the right mouse-button down and drag the pointer to a position some 20 SE of the centre, then release the mouse-button. In some seconds you will see two ellipses appear around the spots.
In the right-hand pane, where the output image is displayed, you can see now an image of KB108 from which the laid lines are suppressed. Now press any key on the keyboard to finish the filtering session - the label 'Filter shape' will change from red to gray - and press the 'Save' button to save the image you filtered.
Note: Whilst the label 'Filter shape' is red you are allowed only to define filter spots on the FFT image. If you finished filtering, or want to change the settings, press any key. You can restart filtering by pressing the 'Filter' button. — If the 'Adjust image' label is red, it shows that the machine is processing the data and that you have to wait until it gets gray again.
2. Enhancing frequencies
By detaching the watermark from the laid lines, you not only are left with a laid-lines free image but also with its complement, an image showing only the laid lines! Select the 'Filtered region > Complement' radiobutton to see them.
Better results can be obtained if you reset the filtered region to 'Selection', enter '3' in the 'Power' field of 'Filter shape', then press the 'Undo' button and reselect the same regions as you have done before. The resulting image will show more clearly the structure of the mould screen, with less noise and more details. Play with the number inputs in 'Power' and see what happens with other values!
You can apply multiple filters, with different powers, on the same image; as well as combining filtershapes - to get for example a ring of power 0, first define a disk of power 0, then in its interior another disk of power 1. Push the 'Show filter' button to see the resulting filter shape and values.
Buttons on top of the BlueNile window let you look closer at the filtering process - please refer to the interface description for details.
You can see that the image you produced with the last filtering is somewhat to uniformly gray. It is possible to accentuate the contrast by selecting the 'Adjust image > Contrast' checkbox. Now it looks better, but there are may points that make it as it where speckled with random pixels - select 'Denoise' to remove this 'salt and pepper' noise. A further step in improving the visual appearence of the image would be to even the background, since you can observe along the vertical chain lines large darker areas. To make the image more uniform, select 'Even'.
'Coloring' has a double functionality. With the FFT displayed with a colormap rather than on a grayscale, it is easier to visualise the topography of the frequency domain image and localise the features you wish to filter. It is also useful for subsequent processing of the filtered image. Displayed with 'Coloring', the watermark of the output image is more uniform than in the input image, showing a greater percentage of red pixels. This is useful as a measure of feature consistency if for example you want to threshold the image and extract the watermark as a white&black image.
The process of selecting regions to be filtered can be automated for some types of images. BlueNile offers you the possibility of filtering automatically and in batches images similar to KB108 that was used here as example. That is images that have features repeated at even intervals across the image. Let's take an example to see how the autofiltering works. Check the 'Autofiltering > Auto' checkbox, set 'Autofiltering > Passes' to 12, deselect all checkboxes that might be selected in 'Adjust image' and put 'Power' to 0. Then load the file eyeraster.jpg that you find in the ..\blunenile\batch\ directory along kb108.jpg.
This image is interesting since it is a largely magnified image scanned from magazine and colors are usually produced by typographical means by superposing a small number of pigments (in this case cyan, magenta, yellow and black). These pigments are deposed on the printed surface as more or less big dots of ink, aligned in evenly spaced rows and columns. This is the typographical raster and there are as many as base colors, rotated at different angles. On the FFT image we clearly see their signatures: four bright 'stars', that form with their symmetrical complements an ellipse around the central spot.
After the image is loaded you will see how BlueNile is selecting 24 spots of high white intensity on the FFT image (double the number of passes you set if 'Filtershape > Symetrical' is on) and that progressively, the output image will look less rasterized. (We might stress that unrasterizing printed material needs specialised software, so don't expect BlueNile to do this. It is howerver interesting that some image improvement can be obtained through an FFT filtering (notice the loss of color brightness and the fact that a denoising leads to similar results).)
A further automatisation is possible by filtering many images one after the other without you having to load them manually. Put the images you want to filter in any free directory, then check 'Auto' and 'Batch' and load any of the images from the directory - BlueNile will filter all of them and save to the same directory.