现在多光谱内配准,然后再和全色配准

cmd/proc.go

@@ -74,7 +74,6 @@ var procCmd = &cobra.Command{
 		if err := reg.DoPhaseCorrelation(); err != nil {
 			logrus.Fatal(err)
 		}
-		reg.DoCoRegistration()
 
 		if params.SaveRegisteredMssRaw {
 			registerdMSSRAW := filepath.Join(
@@ -94,12 +93,6 @@ var procCmd = &cobra.Command{
 		producer.CleanScenes(panScenes)
 		producer.CleanScenes(mssScenes)
 
-		// err := reg.SubScenesLessMem()
-		// if err != nil {
-		// 	logrus.Error(err)
-		// 	return
-		// }
-
 		runtime.GC()
 
 		if saveStrip {

config/config.yaml

@@ -1,3 +1,4 @@
+log:
   log_level: 5
 coregistration:
   mss_bands: 4
@@ -11,8 +12,8 @@ coregistration:
   fus_band_order: "RGB"
 
 radiation:
-  pan_remove_hf_noise: true
+  pan_remove_hf_noise: false
-  mss_remove_hf_noise: true
+  mss_remove_hf_noise: false
   hf_radius_ratio: 0.49
   min_hist_level: 0.3
   max_hist_level: 0.6

pkg/producer/edge_phase_correlation.go

@@ -12,11 +12,14 @@ import (
 // by performing phase correlation on detected
 // edges instead of the raw image
 
-func FindEdges(img gocv.Mat) gocv.Mat {
+func FindEdges(img0 gocv.Mat) gocv.Mat {
-	img0 := gocv.IMRead("/Users/lan/workspace/sjy01/image-proc/data/052022/010/PAN/SJY01_PAN_20240520_115428_052022_103_010_L1A.tiff", gocv.IMReadUnchanged)
 	fmt.Println(img0.Cols(), img0.Rows(), img0.Type().String())
-	img0.ConvertTo(&img0, gocv.MatTypeCV8U)
+	dst8 := gocv.NewMatWithSize(img0.Rows(), img0.Cols(), gocv.MatTypeCV8U)
-	dst := gocv.NewMat()
+	defer dst8.Close()
-	gocv.Canny(img0, &dst, 100, 200)
+	gocv.Normalize(img0, &dst8, 0, 255, gocv.NormMinMax)
-	return dst
+	dst8.ConvertTo(&dst8, gocv.MatTypeCV8U)
+	dstEdge := gocv.NewMat()
+	gocv.Canny(dst8, &dstEdge, 10, 100)
+	dstEdge.ConvertTo(&dstEdge, gocv.MatTypeCV16U)
+	return dstEdge
 }

pkg/producer/image_registration.go

@@ -22,9 +22,9 @@ const (
 	PixelBytes                          = 2
 	PanWidth                            = payload.PAN_PIXEL_WIDTH // 像素宽度
 	MssWidth                            = payload.MSS_PIXEL_WIDTH
-	BlockNH                             = 4
+	BlockNH                             = 8
-	BlockNW                             = 8
+	BlockNW                             = 4
-	OverlappedBlockLines                = 3000 // 重叠块的行数
+	OverlappedBlockLines                = 1000 // 重叠块的行数
 	DownSampled          ResampleMethod = "down_sample_pan"
 	UpSampled            ResampleMethod = "up_sample_mss"
 	PanResolution                       = 1.3 // mm/pixel
@@ -170,12 +170,31 @@ func (r *Registrator) CalcDownPhaseCorrelation() error {
 	blockHeight := r.MssHeight / BlockNH
 	blockWidth := r.MssWidth / BlockNW
 
-	return r.calcPhaseCorrelation(downsampledPanImage, r.MssImages, r.MssHeight, r.MssWidth, blockHeight, blockWidth)
+	// 在 MSS 4 个波段上进行配准
+	err := r.doMSSPhaseCorrelation(r.MssImages[0],
+		[]gocv.Mat{r.MssImages[0], r.MssImages[1], r.MssImages[2], r.MssImages[3]},
+		r.MssHeight, r.MssWidth, blockHeight, blockWidth)
+	if err != nil {
+		return err
+	}
+	r.DoMSSCoRegistration()
 
+	// 基于 PAN 图像进行配准
+	r.phaseShifts[0] = make([]PhaseShiftM, 0)
+	r.deltaXCoeffs[0] = make([]float64, 0)
+	r.deltaYCoeffs[0] = make([]float64, 0)
+	err = r.doMSSPhaseCorrelation(downsampledPanImage,
+		[]gocv.Mat{r.registeredMssImages[0]},
+		r.MssHeight, r.MssWidth, blockHeight, blockWidth)
+	if err != nil {
+		return err
+	}
+	return r.DoPANCoRegistration()
 }
 
 // 将MSS升采样采样后计算相位相关的偏移量
 func (r *Registrator) CalcUpPhaseCorrelation() error {
+	log.Fatal("unsuppotted up-resample method")
 	// 确保 MSS 高度是 PAN 高度的 1/4
 	if r.MssHeight*4 != r.PanHeight {
 		err := fmt.Errorf("MSS height is not 1/4 of PAN height, invalid raw file")
@@ -183,11 +202,19 @@ func (r *Registrator) CalcUpPhaseCorrelation() error {
 		return err
 	}
 
-	// 将PAN将采样作为轮廓匹配基准图像
+	// 在 MSS 4 个波段上进行配准
-	var upsampledMssImages [MssBands]gocv.Mat
+	err := r.doMSSPhaseCorrelation(r.MssImages[0],
+		[]gocv.Mat{r.MssImages[0], r.MssImages[1], r.MssImages[2], r.MssImages[3]},
+		r.MssHeight, r.MssWidth, r.MssHeight/BlockNH, r.MssWidth/BlockNW)
+	if err != nil {
+		return err
+	}
+	r.DoMSSCoRegistration()
+
+	upsampledMssImages := make([]gocv.Mat, MssBands)
 	for i := 0; i < MssBands; i++ {
 		upsampledMssImages[i] = gocv.NewMat()
-		gocv.Resize(r.MssImages[i], &upsampledMssImages[i],
+		gocv.Resize(r.registeredMssImages[i], &upsampledMssImages[i],
 			image.Point{X: r.PanWidth, Y: r.PanHeight}, 0, 0, gocv.InterpolationCubic)
 	}
 
@@ -199,15 +226,23 @@ func (r *Registrator) CalcUpPhaseCorrelation() error {
 
 	log.Infof("blockHeight=%d, blockWidth=%d", blockHeight, blockWidth)
 
-	return r.calcPhaseCorrelation(r.PanImage, upsampledMssImages, r.PanHeight, r.PanWidth, blockHeight, blockWidth)
+	// 基于 PAN 图像进行配准
+	err = r.doMSSPhaseCorrelation(r.PanImage,
+		[]gocv.Mat{upsampledMssImages[0]},
+		r.MssHeight, r.MssWidth, blockHeight, blockWidth)
+	return err
 }
 
-func (r *Registrator) calcPhaseCorrelation(panImage gocv.Mat,
+func (r *Registrator) doMSSPhaseCorrelation(base gocv.Mat,
-	mssImages [4]gocv.Mat,
+	mssImages []gocv.Mat,
 	height, width,
 	blockHeight, blockWidth int) error {
 	var wg sync.WaitGroup
 
+	for band := 0; band < len(mssImages); band++ {
+		r.phaseShifts[band] = make([]PhaseShiftM, 0)
+	}
+
 	for bh := 0; bh < BlockNH; bh++ {
 		for bw := 0; bw < BlockNW; bw++ {
 			wg.Add(1)
@@ -238,8 +273,8 @@ func (r *Registrator) calcPhaseCorrelation(panImage gocv.Mat,
 					x1, y1,
 				)
 
-				panBlock := panImage.Region(rect)
+				panBlock := base.Region(rect)
-				for band := 0; band < MssBands; band++ {
+				for band := 0; band < len(mssImages); band++ {
 					log.Debug("processing band:", band+1, ",block:", bh, rect)
 					mssBlock := mssImages[band].Region(rect)
 
@@ -262,7 +297,7 @@ func (r *Registrator) calcPhaseCorrelation(panImage gocv.Mat,
 
 	wg.Wait()
 
-	for i := 0; i < MssBands; i++ {
+	for i := 0; i < len(mssImages); i++ {
 		for _, shift := range r.phaseShifts[i] {
 			if shift.response > 0.4 || shift.dx > 8 || shift.dy > 8 {
 				log.Debugf("Band %d, block %d, dx=%f, dy=%f, response=%f",
@@ -271,7 +306,7 @@ func (r *Registrator) calcPhaseCorrelation(panImage gocv.Mat,
 		}
 	}
 
-	return r.calcDeltaCoeffs()
+	return r.calcMSSDeltaCoeffs(len(mssImages))
 }
 
 func (r *Registrator) Clean() {
@@ -287,9 +322,9 @@ func (r *Registrator) Clean() {
 	r.rgbirImage.Close()
 }
 
-func (r *Registrator) calcDeltaCoeffs() error {
+func (r *Registrator) calcMSSDeltaCoeffs(bands int) error {
 	// 计算每个通道的delta多项式拟合系数
-	for i := 0; i < MssBands; i++ {
+	for i := 0; i < bands; i++ {
 		var cx []float64
 		var dx []float64
 		var dy []float64
@@ -330,7 +365,7 @@ func (r *Registrator) calcDeltaCoeffs() error {
 		}
 	}
 
-	for i := 0; i < MssBands; i++ {
+	for i := 0; i < bands; i++ {
 		if len(r.deltaXCoeffs[i]) < 2 || len(r.deltaYCoeffs[i]) < 3 {
 			continue
 		}
@@ -344,10 +379,10 @@ func (r *Registrator) calcDeltaCoeffs() error {
 	return nil
 }
 
-func (r *Registrator) DoCoRegistration() error {
+func (r *Registrator) DoMSSCoRegistration() error {
 	for band := 0; band < MssBands; band++ {
 		if len(r.deltaXCoeffs[band]) < 2 || len(r.deltaYCoeffs[band]) < 3 {
-			log.Error("delta coefficients not calculated, skip co-registration")
+			log.Errorf("delta coefficients not calculated, skip co-registration %d", band)
 			r.registeredMssImages[band] = r.MssImages[band].Clone()
 			continue
 		}
@@ -367,12 +402,6 @@ func (r *Registrator) DoCoRegistration() error {
 					dy = r.deltaYCoeffs[band][2]*float64(x*x) + r.deltaYCoeffs[band][1]*float64(x) + r.deltaYCoeffs[band][0] + float64(y)
 				}
 
-				// if band+1 == 4 {
-				// 	fmt.Println("band:", band+1, "x:", x, "map_x:", mx, "y:", y, "map_y:", my)
-				// }
-
-				// mapX.SetFloatAt(y, x, float32(x)+float32(r.deltaXCoeffs[band][0]))
-				// mapY.SetFloatAt(y, x, float32(y)+float32(r.deltaYCoeffs[band][0]))
 				mapX.SetFloatAt(y, x, float32(dx))
 				mapY.SetFloatAt(y, x, float32(dy))
 			}
@@ -392,8 +421,58 @@ func (r *Registrator) DoCoRegistration() error {
 	}
 
 	// 裁掉末尾的的 MSS 480 行 和 PAN 的 480*4 行
-	r.PanHeight -= 480 * 4
+	r.PanHeight -= 360 * 4
-	r.MssHeight -= 480
+	r.MssHeight -= 360
+
+	return nil
+}
+
+func (r *Registrator) DoPANCoRegistration() error {
+	if len(r.deltaXCoeffs[0]) < 2 || len(r.deltaYCoeffs[0]) < 3 {
+		log.Error("delta coefficients not calculated, skip co-registration")
+		return nil
+	}
+
+	mapX := gocv.NewMatWithSize(r.MssHeight, r.MssWidth, gocv.MatTypeCV32FC1)
+	defer mapX.Close()
+	mapY := gocv.NewMatWithSize(r.MssHeight, r.MssWidth, gocv.MatTypeCV32FC1)
+	defer mapY.Close()
+
+	for y := 0; y < r.MssHeight; y++ {
+		for x := 0; x < r.MssWidth; x++ {
+			var dx, dy float64
+			if r.resampleMethod == UpSampled {
+				xx := float64(x * MssBands)
+				yy := float64(y * MssBands)
+				dx = (r.deltaXCoeffs[0][1]*float64(xx) + r.deltaXCoeffs[0][0] + xx) / MssBands
+				dy = (r.deltaYCoeffs[0][2]*float64(xx*xx) + r.deltaYCoeffs[0][1]*float64(xx) + r.deltaYCoeffs[0][0] + yy) / MssBands
+			} else {
+				dx = r.deltaXCoeffs[0][1]*float64(x) + r.deltaXCoeffs[0][0] + float64(x)
+				dy = r.deltaYCoeffs[0][2]*float64(x*x) + r.deltaYCoeffs[0][1]*float64(x) + r.deltaYCoeffs[0][0] + float64(y)
+			}
+
+			mapX.SetFloatAt(y, x, float32(dx))
+			mapY.SetFloatAt(y, x, float32(dy))
+		}
+	}
+
+	log.Println("co-registration for MSS (Align with PAN)")
+	for i := 0; i < MssBands; i++ {
+		log.Infof("r.registeredMssImages[%d]: %v", i, r.registeredMssImages[i].Size())
+		registeredMSS := gocv.NewMatWithSize(r.MssHeight, r.MssWidth, gocv.MatTypeCV16UC1)
+		gocv.Remap(r.registeredMssImages[i],
+			&registeredMSS,
+			&mapX, &mapY,
+			gocv.InterpolationCubic,
+			gocv.BorderConstant,
+			color.RGBA{0, 0, 0, 0})
+
+		r.registeredMssImages[i].Close()
+		r.registeredMssImages[i] = registeredMSS
+	}
+	// 裁掉末尾的的 MSS 480 行 和 PAN 的 480*4 行
+	r.PanHeight -= 120 * 4
+	r.MssHeight -= 120
 
 	return nil
 }

pkg/producer/output.go

@@ -49,6 +49,30 @@ func (r *Registrator) BytesToRaw(mssData []byte, filePath string) error {
 }
 
 func (r *Registrator) SaveRegisteredMssToRaw(raw string) error {
+	return saveRegisteredMssToRaw(raw, r.registeredMssImages)
+}
+
+func (r *Registrator) Report() error {
+	return WriteReport(&r.report, r.Params.ReportFile)
+}
+
+func (r *Registrator) rpcKeywordInTif() {
+	// GDAL库对应的RPC关键词
+	// keys := []string{
+	// 	"ERR_BIAS", "ERR_RAND",
+	// 	"LINE_OFF", "SAMP_OFF",
+	// 	"LAT_OFF", "LONG_OFF", "HEIGHT_OFF",
+	// 	"LINE_SCALE", "SAMP_SCALE",
+	// 	"LAT_SCALE", "LONG_SCALE", "HEIGHT_SCALE",
+	// 	"LINE_NUM_COEFF", "LINE_DEN_COEFF",
+	// 	"SAMP_NUM_COEFF", "SAMP_DEN_COEFF",
+	// }
+
+	// values := map[string]string{}
+
+}
+
+func saveRegisteredMssToRaw(raw string, mssImages [4]gocv.Mat) error {
 	f, err := os.OpenFile(raw, os.O_CREATE|os.O_TRUNC|os.O_WRONLY, 0777)
 	if err != nil {
 		return err
@@ -56,11 +80,11 @@ func (r *Registrator) SaveRegisteredMssToRaw(raw string) error {
 
 	var mssData [4][]byte
 	for i := 0; i < MssBands; i++ {
-		mssData[i] = r.registeredMssImages[i].ToBytes()
+		mssData[i] = mssImages[i].ToBytes()
 	}
 
-	width := r.registeredMssImages[0].Cols() * PixelBytes
+	width := mssImages[0].Cols() * PixelBytes
-	height := r.registeredMssImages[0].Rows()
+	height := mssImages[0].Rows()
 	log.Println("Writing registered MSS to RAW file...", len(mssData[0])*4)
 	log.Println("width:", width)
 	log.Println("height:", height)
@@ -85,23 +109,3 @@ func (r *Registrator) SaveRegisteredMssToRaw(raw string) error {
 
 	return nil
 }
-
-func (r *Registrator) Report() error {
-	return WriteReport(&r.report, r.Params.ReportFile)
-}
-
-func (r *Registrator) rpcKeywordInTif() {
-	// GDAL库对应的RPC关键词
-	// keys := []string{
-	// 	"ERR_BIAS", "ERR_RAND",
-	// 	"LINE_OFF", "SAMP_OFF",
-	// 	"LAT_OFF", "LONG_OFF", "HEIGHT_OFF",
-	// 	"LINE_SCALE", "SAMP_SCALE",
-	// 	"LAT_SCALE", "LONG_SCALE", "HEIGHT_SCALE",
-	// 	"LINE_NUM_COEFF", "LINE_DEN_COEFF",
-	// 	"SAMP_NUM_COEFF", "SAMP_DEN_COEFF",
-	// }
-
-	// values := map[string]string{}
-
-}