To reduce bot traffic you must login to view /lordwelch/---/activity/quarterly.
The GitHub login only links via username.
Files

215 lines
5.0 KiB
Go

package imaging
import (
"image"
"math"
)
type indexWeight struct {
index int
weight float64
}
func precomputeWeights(dstSize, srcSize int, filter ResampleFilter) [][]indexWeight {
du := float64(srcSize) / float64(dstSize)
scale := du
if scale < 1.0 {
scale = 1.0
}
ru := math.Ceil(scale * filter.Support)
out := make([][]indexWeight, dstSize)
tmp := make([]indexWeight, 0, dstSize*int(ru+2)*2)
for v := range dstSize {
fu := (float64(v)+0.5)*du - 0.5
begin := max(int(math.Ceil(fu-ru)), 0)
end := min(int(math.Floor(fu+ru)), srcSize-1)
var sum float64
for u := begin; u <= end; u++ {
w := filter.Kernel((float64(u) - fu) / scale)
if w != 0 {
sum += w
tmp = append(tmp, indexWeight{index: u, weight: w})
}
}
if sum != 0 {
for i := range tmp {
tmp[i].weight /= sum
}
}
out[v] = tmp
tmp = tmp[len(tmp):]
}
return out
}
// Resize resizes the image to the specified width and height using the specified resampling
// filter and returns the transformed image. If one of width or height is 0, the image aspect
// ratio is preserved.
//
// Example:
//
// dstImage := imaging.Resize(srcImage, 800, 600, imaging.Lanczos)
func Resize(img image.Image, width, height int, filter ResampleFilter) *image.NRGBA {
dstW, dstH := width, height
if dstW < 0 || dstH < 0 {
return &image.NRGBA{}
}
if dstW == 0 && dstH == 0 {
return &image.NRGBA{}
}
srcW := img.Bounds().Dx()
srcH := img.Bounds().Dy()
if srcW <= 0 || srcH <= 0 {
return &image.NRGBA{}
}
// If new width or height is 0 then preserve aspect ratio, minimum 1px.
if dstW == 0 {
tmpW := float64(dstH) * float64(srcW) / float64(srcH)
dstW = int(math.Max(1.0, math.Floor(tmpW+0.5)))
}
if dstH == 0 {
tmpH := float64(dstW) * float64(srcH) / float64(srcW)
dstH = int(math.Max(1.0, math.Floor(tmpH+0.5)))
}
// Nearest-neighbor is no longer supported
if srcW != dstW && srcH != dstH {
return resizeVertical(resizeHorizontal(img, dstW, filter), dstH, filter)
}
if srcW != dstW {
return resizeHorizontal(img, dstW, filter)
}
if srcH != dstH {
return resizeVertical(img, dstH, filter)
}
return Clone(img)
}
func resizeHorizontal(img image.Image, width int, filter ResampleFilter) *image.NRGBA {
src := newScanner(img)
dst := image.NewNRGBA(image.Rect(0, 0, width, src.h))
weights := precomputeWeights(width, src.w, filter)
parallel(0, src.h, func(ys <-chan int) {
scanLine := make([]uint8, src.w*4)
for y := range ys {
src.scan(0, y, src.w, y+1, scanLine)
j0 := y * dst.Stride
for x := range weights {
var r, g, b, a float64
for _, w := range weights[x] {
i := w.index * 4
s := scanLine[i : i+4 : i+4]
aw := float64(s[3]) * w.weight
r += float64(s[0]) * aw
g += float64(s[1]) * aw
b += float64(s[2]) * aw
a += aw
}
if a != 0 {
aInv := 1 / a
j := j0 + x*4
d := dst.Pix[j : j+4 : j+4]
d[0] = clamp(r * aInv)
d[1] = clamp(g * aInv)
d[2] = clamp(b * aInv)
d[3] = clamp(a)
}
}
}
})
return dst
}
func resizeVertical(img image.Image, height int, filter ResampleFilter) *image.NRGBA {
src := newScanner(img)
dst := image.NewNRGBA(image.Rect(0, 0, src.w, height))
weights := precomputeWeights(height, src.h, filter)
parallel(0, src.w, func(xs <-chan int) {
scanLine := make([]uint8, src.h*4)
for x := range xs {
src.scan(x, 0, x+1, src.h, scanLine)
for y := range weights {
var r, g, b, a float64
for _, w := range weights[y] {
i := w.index * 4
s := scanLine[i : i+4 : i+4]
aw := float64(s[3]) * w.weight
r += float64(s[0]) * aw
g += float64(s[1]) * aw
b += float64(s[2]) * aw
a += aw
}
if a != 0 {
aInv := 1 / a
j := y*dst.Stride + x*4
d := dst.Pix[j : j+4 : j+4]
d[0] = clamp(r * aInv)
d[1] = clamp(g * aInv)
d[2] = clamp(b * aInv)
d[3] = clamp(a)
}
}
}
})
return dst
}
// ResampleFilter specifies a resampling filter to be used for image resizing.
//
// General filter recommendations:
//
// - Lanczos
// A high-quality resampling filter for photographic images yielding sharp results.
//
// - CatmullRom
// A sharp cubic filter that is faster than Lanczos filter while providing similar results.
//
// - MitchellNetravali
// A cubic filter that produces smoother results with less ringing artifacts than CatmullRom.
//
// - Linear
// Bilinear resampling filter, produces a smooth output. Faster than cubic filters.
//
// - Box
// Simple and fast averaging filter appropriate for downscaling.
// When upscaling it's similar to NearestNeighbor.
//
// - NearestNeighbor
// Fastest resampling filter, no antialiasing.
type ResampleFilter struct {
Support float64
Kernel func(float64) float64
}
// Lanczos filter (3 lobes).
var Lanczos ResampleFilter
func sinc(x float64) float64 {
if x == 0 {
return 1
}
return math.Sin(math.Pi*x) / (math.Pi * x)
}
func init() {
Lanczos = ResampleFilter{
Support: 3.0,
Kernel: func(x float64) float64 {
x = math.Abs(x)
if x < 3.0 {
return sinc(x) * sinc(x/3.0)
}
return 0
},
}
}