Create mask for detected object and calculate Sobel/Laplace only inside it

src/vstarstack/library/data.py

@@ -198,7 +198,7 @@ def get_linked_channel(self, channel : str, link_type : str):
             return None, None, None
         if channel not in self.links[link_type]:
             return None, None, None
-        name = self.links[link_type][channel]
+        name = str(self.links[link_type][channel])
         layer, opts = self.get_channel(name)
         return layer, opts, name
 

src/vstarstack/tool/image_processing/drop_unsharp.py

@@ -26,35 +26,47 @@ class EstimationMethod(Enum):
     SOBEL = 0
     LAPLACE = 1
 
-def measure_sharpness(img : np.ndarray, method : EstimationMethod) -> float:
-    if method == EstimationMethod.SOBEL:
-        sx = scipy.ndimage.sobel(img, axis=0, mode='constant')
-        sy = scipy.ndimage.sobel(img, axis=1, mode='constant')
-        sobel = np.sqrt(sx**2 + sy**2)
-        metric = np.sum(sobel)
-        summ = np.sum(img)
-        return metric / summ
-    elif method == EstimationMethod.LAPLACE:
-        sx = scipy.ndimage.laplace(img, mode='constant')
-        sy = scipy.ndimage.laplace(img, mode='constant')
-        sobel = np.sqrt(sx**2 + sy**2)
-        metric = np.sum(sobel)
-        summ = np.sum(img)
-        return metric / summ
-    else:
-        raise Exception(f"Unknown method {method}")
+def measure_sharpness_sobel(img : np.ndarray, mask : np.ndarray | None) -> float:
+    if mask is not None:
+        img = img * mask
+    sx = scipy.ndimage.sobel(img, axis=0, mode='constant')
+    sy = scipy.ndimage.sobel(img, axis=1, mode='constant')
+    sobel = np.sqrt(sx**2 + sy**2)
+    if mask is not None:
+        sobel = sobel * mask
+    metric = np.sum(sobel)
+    summ = np.sum(img)
+    return metric / summ
+
+def measure_sharpness_laplace(img : np.ndarray, mask : np.ndarray | None) -> float:
+    if mask is not None:
+        img = img * mask
+    sx = scipy.ndimage.laplace(img, mode='constant')
+    sy = scipy.ndimage.laplace(img, mode='constant')
+    laplace = np.sqrt(sx**2 + sy**2)
+    if mask is not None:
+        laplace = laplace * mask
+    metric = np.sum(laplace)
+    summ = np.sum(img)
+    return metric / summ
 
 def measure_sharpness_df(df : vstarstack.library.data.DataFrame, method : EstimationMethod) -> float:
     metric = 0
     nch = 0
+    if method not in [EstimationMethod.LAPLACE, EstimationMethod.SOBEL]:
+        raise Exception(f"Invalid method {method}")
     for channel in df.get_channels():
         img, opts = df.get_channel(channel)
-        if not opts["brightness"]:
+        if not df.get_channel_option(channel, "brightness"):
             continue
         amax = np.amax(img)
         amin = np.amin(img)
         img = (img - amin)/(amax - amin)
-        metric += measure_sharpness(img, method)
+        mask, _, _ = df.get_linked_channel(channel, "mask")
+        if method == EstimationMethod.SOBEL:
+            metric += measure_sharpness_sobel(img, mask)
+        elif method == EstimationMethod.LAPLACE:
+            metric += measure_sharpness_laplace(img, mask)
         nch += 1
     if nch == 0:
         return 0

src/vstarstack/tool/objects/cut.py

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2022 Vladislav Tsendrovskii
+# Copyright (c) 2022-2024 Vladislav Tsendrovskii
 #
 # This program is free software: you can redistribute it and/or modify
 # it under the terms of the GNU General Public License as published by
@@ -14,6 +14,7 @@
 import json
 import os
 import numpy as np
+import cv2
 
 import vstarstack.tool.cfg
 import vstarstack.library.common
@@ -44,10 +45,14 @@ def run(project: vstarstack.tool.cfg.Project, argv: list[str]):
         print(f"Loading info: {name}, r = {r}")
         if r > maxr:
             maxr = r
+    disk_radius=int(maxr+0.5)
     maxr = int(maxr+0.5)+margin
     size = 2*maxr+1
     print("maxr = ", maxr, " size = ", size)
 
+    mask = np.zeros((size,size))
+    cv2.circle(mask, (maxr,maxr), disk_radius, 1, -1)
+
     for name, filename in files:
         print(name)
         with open(filename, encoding='utf8') as f:
@@ -67,13 +72,14 @@ def run(project: vstarstack.tool.cfg.Project, argv: list[str]):
             print("Can not load ", name)
             continue
 
-        weight_links = dict(image.links["weight"])
-
-        for channel in image.get_channels():
+        image.add_channel(mask, "mask", mask=True)
+        for channel in list(image.get_channels()):
             layer, opts = image.get_channel(channel)
-            if opts["encoded"]:
+            if image.get_channel_option(channel, "encoded"):
                 image.remove_channel(channel)
                 continue
+            if image.get_channel_option(channel, "mask"):
+                continue
 
             w = layer.shape[1]
             h = layer.shape[0]
@@ -108,21 +114,19 @@ def run(project: vstarstack.tool.cfg.Project, argv: list[str]):
 
             img = np.zeros((size, size))
             img[target_top:target_top+copy_h, target_left:target_left+copy_w] = layer[source_top:source_top+copy_h, source_left:source_left+copy_w]
+            image.replace_channel(img, channel, **opts)
+            image.add_channel_link(channel, "mask", "mask")
 
             detection["roi"] = {
                 "x1": left,
                 "y1": top,
                 "x2": right,
                 "y2": bottom
             }
-            image.replace_channel(img, channel, **opts)
             vstarstack.tool.common.check_dir_exists(filename)
             with open(filename, "w", encoding='utf8') as f:
                 json.dump(detection, f, indent=4, ensure_ascii=False)
 
-        for ch, value in weight_links.items():
-            image.add_channel_link(ch, value, "weight")
-
         image.params["w"] = size
         image.params["h"] = size