97 lines
2.7 KiB
Go
97 lines
2.7 KiB
Go
package rrc
|
||
|
||
import (
|
||
"math"
|
||
|
||
log "github.com/sirupsen/logrus"
|
||
"gocv.io/x/gocv"
|
||
)
|
||
|
||
// Destriping multisensor imagery with moment matching [Gadallah, 2000]
|
||
func DoMomentMatching(originalImg gocv.Mat) {
|
||
probes := originalImg.Cols()
|
||
log.Printf("do moment matching for %d probes, %d rows", probes, originalImg.Rows())
|
||
// 第i个探元的像元均值和标准差
|
||
means := make([]float64, probes)
|
||
stds := make([]float64, probes)
|
||
|
||
// 计算每个探元的均值和标准差
|
||
for x := 0; x < originalImg.Cols(); x++ {
|
||
var total int64
|
||
n := originalImg.Rows()
|
||
var dn uint16
|
||
for y := 0; y < n; y++ {
|
||
dn = uint16(originalImg.GetShortAt(y, x))
|
||
total += int64(dn)
|
||
}
|
||
means[x] = float64(total) / float64(n)
|
||
|
||
var a float64
|
||
for y := 0; y < n; y++ {
|
||
dn = uint16(originalImg.GetShortAt(y, x))
|
||
a += math.Pow(float64(dn)-means[x], 2)
|
||
}
|
||
stds[x] = math.Sqrt(a / float64(n))
|
||
}
|
||
|
||
// 列参考值和列参考标准差
|
||
var mu, sig float64
|
||
for x := 0; x < probes; x++ {
|
||
mu += means[x]
|
||
sig += stds[x]
|
||
}
|
||
|
||
mu = mu / float64(probes)
|
||
sig = sig / float64(probes)
|
||
|
||
log.Printf("mean reference value: %f, std reference value: %f", mu, sig)
|
||
|
||
// 修正 DN_adjusted[i] = (DN[i] - means[i]) *sig/stds[i]+mu
|
||
for x := 0; x < originalImg.Cols(); x++ {
|
||
for y := 0; y < originalImg.Rows(); y++ {
|
||
dn := uint16(originalImg.GetShortAt(y, x))
|
||
dn_adjusted := uint16((float64(dn)-means[x])*sig/stds[x] + mu)
|
||
originalImg.SetShortAt(y, x, int16(dn_adjusted))
|
||
}
|
||
}
|
||
}
|
||
|
||
// 相邻列均衡算法（张兵，2006）- 实测结果不如上面的矩匹配算法
|
||
func DoMomentMatching2006(originalImg gocv.Mat) {
|
||
probes := originalImg.Cols()
|
||
log.Printf("do moment matching for %d probes", probes)
|
||
// 第i个探元的像元均值和标准差
|
||
means := make([]float64, probes)
|
||
stds := make([]float64, probes)
|
||
// 计算每个探元的均值和标准差
|
||
for x := 0; x < originalImg.Cols(); x++ {
|
||
var total int64
|
||
n := originalImg.Rows()
|
||
var dn uint16
|
||
for y := 0; y < n; y++ {
|
||
dn = uint16(originalImg.GetShortAt(y, x))
|
||
total += int64(dn)
|
||
}
|
||
means[x] = float64(total) / float64(n)
|
||
|
||
var a float64
|
||
for y := 0; y < n; y++ {
|
||
dn = uint16(originalImg.GetShortAt(y, x))
|
||
a += math.Pow(float64(dn)-means[x], 2)
|
||
}
|
||
stds[x] = math.Sqrt(a / float64(n))
|
||
}
|
||
|
||
// 修正 DN_adjusted[i] = (DN[i] - means[i]) *sig/stds[i]+mu
|
||
for x := 1; x < originalImg.Cols()-1; x++ {
|
||
// 列参考值和列参考标准差
|
||
mu := (means[x-1] + means[x+1]/2 + means[x]) / 2
|
||
sig := (stds[x-1] + stds[x+1]/2 + stds[x]) / 2
|
||
for y := 0; y < originalImg.Rows(); y++ {
|
||
dn := uint16(originalImg.GetShortAt(y, x))
|
||
dn_adjusted := uint16((float64(dn)-means[x])*sig/stds[x] + mu)
|
||
originalImg.SetShortAt(y, x, int16(dn_adjusted))
|
||
}
|
||
}
|
||
}
|