mirror of
https://github.com/hajimehoshi/ebiten.git
synced 2025-01-26 10:42:42 +01:00
examples/blur: fix the blur filter example to be a real box blur. (#2000)
Also add a note that this is not how games should implement a blur if avoidable. Krita's blur and the original blur: https://rm.cloudns.org/img/uploaded/29c0418f25bf56b802f746d2f269e215.png The original image and the new blur (Krita's): https://rm.cloudns.org/img/uploaded/bd04828baf6c01fcf7301a6672a4c406.png Note how Ebiten's blur has almost no vertical component. Note how after the change it looks basically like Krita's.
This commit is contained in:
parent
62229e82a6
commit
99aef33970
@ -49,17 +49,39 @@ func (g *Game) Draw(screen *ebiten.Image) {
|
|||||||
|
|
||||||
// Box blur (7x7)
|
// Box blur (7x7)
|
||||||
// https://en.wikipedia.org/wiki/Box_blur
|
// https://en.wikipedia.org/wiki/Box_blur
|
||||||
|
//
|
||||||
|
// Note that this is a fixed function implementation of a box blur - more
|
||||||
|
// efficiency can be gained by using a separable blur
|
||||||
|
// (blurring horizontally and vertically separately, or for large blurs,
|
||||||
|
// even multiple horizontal or vertical passes), ideally combined with
|
||||||
|
// doing the summing up in a fragment shader (Kage can be used here).
|
||||||
|
//
|
||||||
|
// So this implementation only serves to demonstrate use of alpha blending.
|
||||||
|
layers := 0
|
||||||
for j := -3; j <= 3; j++ {
|
for j := -3; j <= 3; j++ {
|
||||||
for i := -3; i <= 3; i++ {
|
for i := -3; i <= 3; i++ {
|
||||||
op := &ebiten.DrawImageOptions{}
|
op := &ebiten.DrawImageOptions{}
|
||||||
op.GeoM.Translate(float64(i), 244+float64(j))
|
op.GeoM.Translate(float64(i), 244+float64(j))
|
||||||
// Alpha scale should be 1.0/49.0, but accumulating 1/49 49 times doesn't reach to 1, because
|
// This is a blur based on the CompositerModeSourceOver composition mode,
|
||||||
// the final color is affected by the destination alpha when CompositeModeSourceOver is used.
|
// which is basically (GL_ONE, GL_ONE_MINUS_SRC_ALPHA). ColorM acts
|
||||||
|
// on unpremultiplied colors, but all Ebiten internal colors are
|
||||||
|
// premultiplied, meaning this mode is regular alpha blending,
|
||||||
|
// computing each destination pixel as srcPix * alpha + dstPix * (1 - alpha).
|
||||||
|
//
|
||||||
|
// This means that the final color is affected by the destination color when CompositeModeSourceOver is used.
|
||||||
// This composite mode is the default mode. See how this is calculated at the doc:
|
// This composite mode is the default mode. See how this is calculated at the doc:
|
||||||
// https://pkg.go.dev/github.com/hajimehoshi/ebiten/v2#CompositeMode
|
// https://pkg.go.dev/github.com/hajimehoshi/ebiten/v2#CompositeMode
|
||||||
//
|
//
|
||||||
// Use a higher value than 1.0/49.0. Here, 1.0/25.0 here to get a reasonable result.
|
// So if using the same alpha every time, the end result will sure be biased towards the last layer.
|
||||||
op.ColorM.Scale(1, 1, 1, 1.0/25.0)
|
//
|
||||||
|
// Correct averaging works based on
|
||||||
|
// Let A_n := (a_1 + ... + a_n) / n
|
||||||
|
// A_{n+1} = (a_1 + ... + a_{n+1}) / (n + 1)
|
||||||
|
// A_{n+1} = (n * A_n + a_{n+1)) / (n + 1)
|
||||||
|
// A_{n+1} = A_n * (1 - 1/(n+1)) + a_{n+1} * 1/(n+1)
|
||||||
|
// which is precisely what an alpha blend with alpha 1/(n+1) does.
|
||||||
|
layers++
|
||||||
|
op.ColorM.Scale(1, 1, 1, 1.0/float64(layers))
|
||||||
screen.DrawImage(gophersImage, op)
|
screen.DrawImage(gophersImage, op)
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
Loading…
Reference in New Issue
Block a user