DFT 低通滤波器

pkg/rrc/filter_test.go

@@ -0,0 +1,43 @@
+package rrc
+
+import (
+	"testing"
+
+	"github.com/airbusgeo/godal"
+	"gocv.io/x/gocv"
+	"starwiz.cn/sjy01/image-proc/pkg/utils"
+)
+
+func TestFilter(t *testing.T) {
+	godal.RegisterAll()
+	tif := "/Users/lan/workspace/sjy01/image-proc/data/052022-no-rrc/010/PAN/SJY01_PAN_20240520_115428_052022_103_010_L1A.tiff"
+	tif = "/Users/lan/workspace/sjy01/image-proc/data/051622/007/PAN/SJY01_PAN_20240516_101236_051622_096_007_L1A.tiff"
+	// tif = "/Users/lan/workspace/sjy01/image-proc/data/060622-mm/003/PAN/SJY01_PAN_20240606_012004_060622_065_003_L1A.tiff"
+
+	ds, err := godal.Open(tif)
+	if err != nil {
+		t.Error(err)
+	}
+	bands := ds.Bands()
+	width := bands[0].Structure().SizeX
+	height := bands[0].Structure().SizeY
+	img := gocv.NewMatWithSize(height, width, gocv.MatTypeCV16UC1)
+
+	data := make([]uint16, width*height)
+	err = bands[0].Read(0, 0, data, width, height)
+	if err != nil {
+		t.Error(err)
+	}
+
+	// 将 16 位数据存储到图像的相应通道
+	for y := 0; y < height; y++ {
+		for x := 0; x < width; x++ {
+			value := data[y*width+x]
+			img.SetShortAt(y, x, int16(value))
+		}
+	}
+
+	m0 := HFNoiseFilter(img, float64(img.Cols())*0.45)
+	defer m0.Close()
+	utils.SavePanToGDALGTiff(m0, 0, 0, "/Users/lan/workspace/sjy01/image-proc/data/fft/filtered.tiff", 1.3)
+}

pkg/rrc/hf_filter.go

@@ -0,0 +1,118 @@
+package rrc
+
+import (
+	"image"
+	"math"
+
+	"github.com/sirupsen/logrus"
+	"gocv.io/x/gocv"
+)
+
+// filter high frequence noise
+func HFNoiseFilter(img gocv.Mat, radius float64) gocv.Mat {
+	logrus.Println("applying hf noise filter...")
+	imgFloat := gocv.NewMat()
+	img.ConvertTo(&imgFloat, gocv.MatTypeCV32F)
+
+	// rows, cols := img.Rows(), img.Cols()
+	// utils.SavePanToGDALGTiff(img, 0, 0, "/Users/lan/workspace/sjy01/image-proc/data/fft/original.tiff", 1.3)
+
+	// 将图像转换为32位浮点型
+	img32f := gocv.NewMat()
+	img.ConvertTo(&img32f, gocv.MatTypeCV32F)
+	defer img32f.Close()
+
+	// 获取图像尺寸
+	rows, cols := img32f.Rows(), img32f.Cols()
+
+	// zeros := gocv.NewMatFromScalar(gocv.NewScalar(0, 0, 0, 0), gocv.MatTypeCV32F)
+
+	// 执行 DFT
+	DFT := gocv.NewMat()
+	defer DFT.Close()
+
+	planes := gocv.NewMatWithSize(rows, cols, gocv.MatTypeCV32F)
+	img32f.ConvertTo(&planes, gocv.MatTypeCV32F)
+	gocv.Merge([]gocv.Mat{planes, gocv.Zeros(rows, cols, gocv.MatTypeCV32F)}, &DFT)
+	gocv.DFT(DFT, &DFT, gocv.DftComplexOutput)
+	fftshift(&DFT)
+	// 将复数部分分离并计算振幅光谱
+	pss := gocv.Split(DFT)
+	// SaveDFT(DFT, "/Users/lan/workspace/sjy01/image-proc/data/fft/dft.tiff")
+
+	// 应用滤波器
+	for y := 0; y < rows; y++ {
+		for x := 0; x < cols; x++ {
+			if math.Sqrt(math.Pow(float64(x-cols/2), 2)+math.Pow(float64(y-rows/2), 2)) > float64(radius) {
+				// if math.Abs(float64(x-cols/2)) < float64(radius) || math.Abs(float64(y-rows/2)) < float64(radius) {
+				pss[0].SetFloatAt(y, x, 0.0)
+				pss[1].SetFloatAt(y, x, 0.0)
+			}
+		}
+	}
+	gocv.Merge(pss, &DFT)
+	// SaveDFT(DFT, "/Users/lan/workspace/sjy01/image-proc/data/fft/dft_filtered.tiff")
+	fftshift(&DFT)
+	// 执行 IDFT
+	IDFT := gocv.NewMat()
+	gocv.DFT(DFT, &IDFT, gocv.DftInverse|gocv.DftRealOutput)
+
+	// 归一化到0-65535
+	gocv.Normalize(IDFT, &IDFT, 0, 8192, gocv.NormMinMax)
+	IDFT.ConvertTo(&IDFT, gocv.MatTypeCV16U)
+
+	// 保存 IDFT 结果
+	return IDFT
+}
+
+// 创建低通滤波器
+func createLowPassFilter(filter *gocv.Mat, radius int) {
+	rows, cols := filter.Rows(), filter.Cols()
+	center := image.Pt(cols/2, rows/2)
+	for y := 0; y < rows; y++ {
+		for x := 0; x < cols; x++ {
+			dist := math.Sqrt(math.Pow(float64(x-center.X), 2) + math.Pow(float64(y-center.Y), 2))
+			if dist <= float64(radius) {
+				filter.SetFloatAt(y, x, 1.0)
+			} else {
+				filter.SetFloatAt(y, x, 0.0)
+			}
+		}
+	}
+}
+
+// fftshift 将频率直流分量移动到图像中心
+func fftshift(input *gocv.Mat) {
+	cx, cy := input.Cols()/2, input.Rows()/2
+
+	q0 := input.Region(image.Rect(0, 0, cx, cy))                       // Top-Left
+	q1 := input.Region(image.Rect(cx, 0, input.Cols(), cy))            // Top-Right
+	q2 := input.Region(image.Rect(0, cy, cx, input.Rows()))            // Bottom-Left
+	q3 := input.Region(image.Rect(cx, cy, input.Cols(), input.Rows())) // Bottom-Right
+
+	tmp := gocv.NewMatWithSize(q0.Rows(), q0.Cols(), input.Type())
+	q0.CopyTo(&tmp)
+	q3.CopyTo(&q0)
+	tmp.CopyTo(&q3)
+
+	q1.CopyTo(&tmp)
+	q2.CopyTo(&q1)
+	tmp.CopyTo(&q2)
+}
+
+func SaveDFT(DFT gocv.Mat, file string) {
+	pss := gocv.Split(DFT)
+	planes := gocv.NewMat()
+	gocv.Magnitude(pss[0], pss[1], &planes)
+	// 进行对数尺度变换，便于可视化
+	ones := gocv.NewMatFromScalar(gocv.NewScalar(1, 0, 0, 0), gocv.MatTypeCV64F)
+	gocv.Add(planes, ones, &planes)
+	gocv.Log(planes, &planes)
+
+	// 归一化
+	gocv.Normalize(planes, &planes, 0, 1, gocv.NormMinMax)
+
+	// 保存DFT结果
+	gocv.IMWrite(file, planes)
+
+}

pkg/rrc/hf_filter_fft.go

@@ -0,0 +1,56 @@
+package rrc
+
+import (
+	"math"
+
+	"github.com/mjibson/go-dsp/fft"
+	log "github.com/sirupsen/logrus"
+	"gocv.io/x/gocv"
+)
+
+func HFNoiseFilterFFT(input gocv.Mat, radius int) gocv.Mat {
+	// 将灰度图像转换为浮点数数组
+	height := input.Rows()
+	width := input.Cols()
+
+	data := make([][]complex128, height)
+	for y := 0; y < height; y++ {
+		data[y] = make([]complex128, width)
+		for x := 0; x < width; x++ {
+			grayColor := float64(uint16(input.GetShortAt(y, x)))
+			data[y][x] = complex(grayColor, 0)
+		}
+	}
+
+	// 进行二维DFT
+	log.Println("Performing 2D DFT...")
+	dft := fft.FFT2(data)
+	// 应用低通滤波器
+	log.Println("Applying low-frequency-pass filter...")
+	for y := 0; y < height; y++ {
+		for x := 0; x < width; x++ {
+			if math.Abs(float64(x-width/2)) < float64(radius) || math.Abs(float64(y-height/2)) < float64(radius) {
+				dft[y][x] = 0
+			}
+		}
+	}
+
+	// 进行二维逆DFT
+	log.Println("Performing 2D inverse DFT...")
+	idft := fft.IFFT2(dft)
+
+	// 将浮点数数组转换回灰度图像
+	for y := 0; y < height; y++ {
+		for x := 0; x < width; x++ {
+			realPart := real(idft[y][x])
+			if realPart < 0 {
+				realPart = 0
+			} else if realPart > 65535 {
+				realPart = 65535
+			}
+			input.SetShortAt(y, x, int16(uint16(realPart)))
+		}
+	}
+
+	return input
+}

pkg/rrc/mean_filter.go

@@ -0,0 +1,15 @@
+package rrc
+
+import (
+	"image"
+
+	log "github.com/sirupsen/logrus"
+	"gocv.io/x/gocv"
+)
+
+func MeanFilter(img gocv.Mat, ksize int) gocv.Mat {
+	log.Println("Applying mean filter, kernel size:", ksize)
+	meanFilteredImg := gocv.NewMat()
+	gocv.Blur(img, &meanFilteredImg, image.Pt(ksize, ksize))
+	return meanFilteredImg
+}