Downsample+flood fill algorithm (it almost works!)

1 files changed, 130 insertions, 14 deletions

diff --git a/decoder.py b/decoder.py
index e58e246..6606b03 100644
--- a/decoder.py
+++ b/decoder.py
@@ -1,5 +1,7 @@
 import argparse
+import collections
 import cv2
+import matplotlib.pyplot as plt
 import numpy as np
 from creedsolo import RSCodec
 from raptorq import Decoder
@@ -14,34 +16,148 @@ parser.add_argument("--level", help="error correction level", default=0.1, type=
 parser.add_argument("--device", help="camera device index", default=1, type=int)
 args = parser.parse_args()
 
-cheight = args.height // 10
-cwidth = args.width // 10
+cheight = cwidth = max(args.height // 10, args.width // 10)
 frame_size = args.height * args.width - 4 * cheight * cwidth
 frame_xor = np.arange(frame_size, dtype=np.uint8)
-rs_size = int(frame_size * (1 - args.level))
+rs_size = frame_size - int((frame_size + 254) / 255) * int(args.level * 255) - 4
 
-rsc = RSCodec(frame_size - rs_size)
-raptor_decoder = Decoder.with_defaults(args.len, rs_size)
+rsc = RSCodec(int(args.level * 255))
+decoder = Decoder.with_defaults(args.len, rs_size)
 data = None
-cap = cv2.VideoCapture(args.device)
+# cap = cv2.VideoCapture(args.device)
 while data is None:
-    ret, raw_frame = cap.read()
-    if not ret:
-        continue
-    color_frame = decode(raw_frame)  # TODO
+    # ret, raw_frame = cap.read()
+    # if not ret:
+    #    continue
+    raw_frame = cv2.cvtColor(
+        cv2.imread("/home/a/Pictures/Camera/IMG_20240422_000849_027.jpg"),
+        cv2.COLOR_BGR2RGB,
+    ).astype(np.float64)
+
+    X, Y = raw_frame.shape[:2]
+    scale = min(X // 20, Y // 20)
+    # Resize so smaller dim is 20
+    # Use fast default interpolation for factor of 4
+    # Then switch to good slow interpolation
+    dframe = cv2.resize(
+        cv2.resize(raw_frame, (Y // 4, X // 4)),
+        (Y // scale, X // scale),  # OpenCV swaps them
+        interpolation=cv2.INTER_AREA,
+    )
+    plt.imshow(dframe.astype(np.uint8))
+    plt.show()
+
+    def max_in_orig(x):
+        return tuple(
+            np.array(np.unravel_index(np.argmax(x), x.shape)) * scale + scale // 2
+        )
+
+    sumframe = np.sum(dframe, axis=2)
+    # TODO: Only search in corner area
+    widx = max_in_orig((np.std(dframe, axis=2) < 35) * sumframe)
+    ridx = max_in_orig(2 * dframe[:, :, 0] - sumframe)
+    gidx = max_in_orig(2 * dframe[:, :, 1] - sumframe)
+    bidx = max_in_orig(2 * dframe[:, :, 2] - sumframe)
+
+    # Flood fill corners
+    def flood_fill(s):
+        vis = np.full((X, Y), False)
+        vis[s] = True
+        queue = collections.deque([s])
+        pos = np.array(s)
+        col = np.copy(raw_frame[s])
+        n = 1
+        while len(queue) > 0:
+            u = queue.popleft()
+            for d in [(5, 0), (0, 5), (-5, 0), (0, -5)]:
+                v = (u[0] + d[0], u[1] + d[1])
+                if (
+                    0 <= v[0] < X
+                    and 0 <= v[1] < Y
+                    and not vis[v]
+                    and np.linalg.norm(raw_frame[v] - raw_frame[s]) < 100
+                ):
+                    vis[v] = True
+                    pos += np.array(v)
+                    col += raw_frame[v]
+                    n += 1
+                    queue.append(v)
+        plt.imshow(raw_frame.astype(np.uint8))
+        plt.scatter(*reversed(np.where(vis)))
+        plt.scatter(pos[1] / n, pos[0] / n)
+        plt.show()
+        return pos / n, col / n
+
+    widx, wcol = flood_fill(widx)
+    ridx, rcol = flood_fill(ridx)
+    gidx, gcol = flood_fill(gidx)
+    bidx, bcol = flood_fill(bidx)
+
+    # Find basis of color space
+    origin = (rcol + gcol + bcol - wcol) / 2
+    rcol -= origin
+    gcol -= origin
+    bcol -= origin
+    print(origin, rcol, gcol, bcol)
+    F = 255 * np.linalg.inv(np.stack((rcol, gcol, bcol)).T)
+
+    # Dumb perspective transform
+    xv = np.linspace(
+        -(cheight / 2 - 1) / (args.height - cheight + 1),
+        1 + (cheight / 2 - 1) / (args.height - cheight + 1),
+        args.height,
+    )
+    yv = np.linspace(
+        -(cwidth / 2 - 1) / (args.width - cwidth + 1),
+        1 + (cwidth / 2 - 1) / (args.width - cwidth + 1),
+        args.width,
+    )
+    xp = (
+        np.outer(1 - xv, 1 - yv) * widx[0]
+        + np.outer(1 - xv, yv) * ridx[0]
+        + np.outer(xv, 1 - yv) * gidx[0]
+        + np.outer(xv, yv) * bidx[0]
+    )
+    yp = (
+        np.outer(1 - xv, 1 - yv) * widx[1]
+        + np.outer(1 - xv, yv) * ridx[1]
+        + np.outer(xv, 1 - yv) * gidx[1]
+        + np.outer(xv, yv) * bidx[1]
+    )
+
+    plt.scatter(widx[1], widx[0])
+    plt.scatter(ridx[1], ridx[0])
+    plt.scatter(gidx[1], gidx[0])
+    plt.scatter(bidx[1], bidx[0])
+    plt.scatter(yp, xp)
+    plt.imshow(raw_frame.astype(np.uint8))
+    plt.show()
+    print(111111111, xp)
+    print(111111111, yp)
+
+    raw_color_frame = raw_frame[xp.astype(np.int64), yp.astype(np.int64), :]
+    print(raw_color_frame)
+    color_frame = np.clip(
+        np.squeeze(F @ (raw_color_frame - origin)[..., np.newaxis]), 0, 255
+    ).astype(np.uint8)
+    print(color_frame)
     frame = (
-        (color_frame[:, :, 0] >> 5 & 0b00000111)
-        + (color_frame[:, :, 1] >> 2 & 0b00111000)
+        (color_frame[:, :, 0] >> 5)
+        + (color_frame[:, :, 1] >> 3 & 0b00111000)
         + (color_frame[:, :, 2] & 0b11000000)
     )
     frame_data = np.concatenate(
         (
             frame[:cheight, cwidth : args.width - cwidth].flatten(),
             frame[cheight : args.height - cheight].flatten(),
-            frame[args.heigth - cheight, cwidth : args.width - cwidth].flatten(),
+            frame[args.height - cheight :, cwidth : args.width - cwidth].flatten(),
         )
     )
-    data = raptor_decoder.decode(rsc.decode(frame_data ^ frame_xor))
+    print(list(frame_data))
+    tmp = rsc.decode(frame_data ^ frame_xor)
+    # print(tmp, list(tmp[2]), bytes(tmp[0]))
+    data = decoder.decode(bytes(tmp))
+    break
 with open(args.file, "wb") as f:
     f.write(data)
 cap.release()