Resized from 1000×1000 to 120×120 using pixel mixing.
IW handles this 250,000× size reduction correctly.
Nearest-neighbor resizing. Not recommended.
A Hermite filter.
A Mitchell filter.
A 3-lobed lanczos filter.
An 8-lobed lanczos filter. This has some nice theoretical properties, bit it gives poor results with any real-world image that has sharp edges.
A gaussian filter. This is too blurry for most purposes.
Gaussian blur. A gaussian filter is a natural filter to use for blurring. Certain other filters such as bspline and hermite also give good results. Filters that take negative values, such lanczos and mitchell, may give poor results when used for blurring.
Blurred only in the horizontal direction.
Blurred only in the vertical direction.
Convert to grayscale.
Suboptimal grayscale conversion as done by many applications. Some of the bright colors are too dark.
Channel offset.
Posterize to 8 colors.
Posterize to 64 colors.
Floyd-Steinberg dithering to an 8-color color cube.
Dithering with incorrect gamma correction. This makes the image too light.
Sierra dither.
An ordered dither.
A halftone dither.
Random dithering, using different thresholds for each channel.
Random dithering, using the same thresholds for all channels.
3 color levels per channel (27 colors).
6 color levels per channel (216 colors).
Dither only the red channel.
Dither only the green channel.
Dither only the blue channel.
Dither the red channel with Floyd-Steinberg dithering, and the green channel with ordered dithering.
A bilevel (black and white; 2-color) image.
Grayscale with 4 gray levels, using a linear colorspace.
Grayscale with 4 gray levels, using a default colorspace (approx. gamma=2.2). A different set of 4 gray levels are available than in the previous image. IW knows which levels are available, and takes that into account.
Grayscale with 4 gray levels, using a gamma=3.5 colorspace. A different set of 4 gray levels are available than in the previous image two images.
