Return to the manual's home page.
filter.c gives a list of five filters types : hue_filter sat_filter val_filter contrast_ gradient_
The first thres types are "simple", one could say "of the first order": the deal with each pixel independently of its neighbours as well as of the image as a whole. The other types take into account, more or less, the environment.
hue_filter.c list the filtres operating mainly on the hues
negative (or shortly n) inverses values and gives the complementary color
hue_rotation operates (would do that if it worked correctly) a rotation in the hue space (see the chapter about colors), indicated by the number of degrees, for example 90 hue_rotation. Vous pouvez apprécier l'effet sur une image normale, mais on voit directemnet l'effet angulaire en chargeant color_wheel.
about BW, graying and smart_graying see the chapter about colors/
- saturation is slightly reduced by sat- more strongly with sat-- . You can trim the effet with a number followed by sat-num (maximum effect with 0 sat-num, equivalent to graying)
- saturation is slightly augmented by sat+, more strongly with sat++ ; and you can adjust as above with sat+num (maximum effect wih 1000 sat+num)
val filter.c change the luminosity with darken darken+ lighten lighten+
contrast augments contrast
More complex filters
fuzz makes a light fuzzing; you can repeat the operation for example as a grassroot form of antialiasing. This operations does, on each pixel, the average with the neighbouring pixels.
ranfuzz is not formally a fuzzing algorithm. It replaces each pixel by another pixel in the images, drawn at random in a more or less wide square. The results are pleasant. It can be parametered by a number given before (10 by default); this number gives the size of the square side.
ranfuzzh and ranfuzzv apply the same operation, but only horizontally or vertically.
median also create some fuzziness, but more subtly: it takes for each pixel the median value in the rectangle surrounding this pixel. On simple images (got by draw square, for instance) the result if of little interest. But on real pictures, or mixing of pictures, if makes sometimes a very pleasant merging. It can be of use also before a segmentation.
All the filters above tend to smooth the image, to melt th pixels. The following ones, on the other hand, stenghten the differences;
sharpen gives a "sharper", "crisper" image. You can use it repeatedly, but the results become quire rapidly bizarre.
gradient replaces the images by the variations inside it; the result by itself is rarely pleasant, but gives useful bases tor other actinos. You can make it better with an additional sharpen darken
contours, as the name says, reduces the images to conturs. The result on a natural image is not always pleasant. Try, on a flat bakground paint square contours . Sometimes you can enhance the results with the morphogenetic fuctions. Or by preparing the image by median or by a segmentation followed by segapla (whic dyes each region with the average color in this region). Beware, that takes time.
Some geometric operations (see next chapter) can also be considred as filters.