commit operation

gravure_upic/circular_to_plane/circular_to_plain.py

@@ -0,0 +1,26 @@
+import cv2
+import math
+
+ring = cv2.imread('../circular.png')
+
+size = ring.shape[0]
+outer_radius = size // 2
+inner_radius = 0
+
+unwrapped = cv2.warpPolar(
+    ring,
+    (size, int(size * math.pi)),
+    (outer_radius, outer_radius),
+    outer_radius,
+    flags = 0
+)
+
+unwrapped = cv2.rotate(unwrapped, cv2.ROTATE_90_COUNTERCLOCKWISE)
+
+unwrapped = unwrapped[inner_radius:, :]
+
+cv2.imshow("original", ring)
+cv2.imshow("Unwrapped", unwrapped)
+cv2.waitKey(0)
+cv2.destroyAllWindows()
+cv2.imwrite("output.png", unwrapped)

gravure_upic/img_to_freq/img_to_freq.py

@@ -0,0 +1,101 @@
+from tqdm import tqdm
+from scipy.ndimage import uniform_filter1d
+from scipy.io.wavfile import write
+import math, cv2, sys, getopt, wave
+import numpy as np
+
+def unwrap_img(input_file):
+    ring_img = cv2.imread(input_file)
+
+    size = ring_img.shape[0]
+    outer_radius = size // 2
+    inner_radius = 0
+
+    unwrapped_img = cv2.warpPolar(
+        ring_img,
+        (size, int(size * math.pi)),
+        (outer_radius, outer_radius),
+        outer_radius,
+        flags=0
+    )
+
+    rotated_unwrapped = cv2.rotate(unwrapped_img, cv2.ROTATE_90_COUNTERCLOCKWISE)
+    cropped_unwrapped = rotated_unwrapped[inner_radius:, :]
+
+    inverted_unwrapped = cv2.bitwise_not(cropped_unwrapped)
+
+    return inverted_unwrapped
+
+
+def image_to_audio(input_img, out_wav, duration_seconds, sample_rate, vertical_res, amp_threshold):
+    max_freq = 10000
+    min_freq = 50
+
+    duration_seconds = float(duration_seconds)
+
+    # Downsample image vertically to reduce number of frequencies
+    input_img = input_img[::vertical_res]  # Use one row every 20 pixels
+    height, width, _ = input_img.shape
+
+    num_samples = int(sample_rate * duration_seconds)
+    freqs = np.logspace(np.log10(min_freq), np.log10(max_freq), height)[::-1]
+
+    brightness = np.mean(input_img / 255.0, axis=2)
+    amplitudes = np.where(brightness >= 0.1, brightness, 0)
+
+    samples = np.zeros(num_samples, dtype=np.float32)
+    chunk_size = 10000
+
+    for start in tqdm(range(0, num_samples, chunk_size)):
+        end = min(start + chunk_size, num_samples)
+        t = np.linspace(start / sample_rate, end / sample_rate, end - start)
+
+        pixel_xs = (t * width / duration_seconds).astype(int)
+        pixel_xs = np.clip(pixel_xs, 0, width - 1)
+
+        amp_per_sample = amplitudes[:, pixel_xs]
+        amp_per_sample[amp_per_sample < amp_threshold] = 0
+        amp_per_sample = uniform_filter1d(amp_per_sample, size=7, axis=1)  # Smooth
+        
+        phases = 2 * np.pi * freqs[:, None] * t[None, :]
+        active = np.count_nonzero(amp_per_sample, axis=0)
+        chunk = np.sum(amp_per_sample * np.sin(phases), axis=0)
+
+        # Normalize chunk by number of active oscillators to avoid "snow"
+        chunk = np.where(active > 0, chunk / active, 0)
+        samples[start:end] = chunk
+
+    # Final normalization
+    samples /= np.max(np.abs(samples) + 1e-8)
+
+    # Convert to int16 for wav
+    wav_samples = (samples * 32767).astype(np.int16)
+    write(out_wav, sample_rate, wav_samples)
+    print(f"Audio saved to {out_wav}")
+
+
+if __name__ == '__main__':
+    input_file = ''
+    output_file = ''
+    duration = 10
+
+    try:
+        opts, args = getopt.getopt(sys.argv[1:], "hi:o:d:yRes:thresh:")
+    except getopt.GetoptError:
+        print('error: img_to_freq.py -i <input_picture> -o <output_sound> -d <audio_duration>')
+
+    for opt, arg in opts:
+        if opt == '-h':
+            print('error: img_to_freq.py -i <input_picture> -o <output_sound> -d <audio_duration>')
+            sys.exit(2)
+        elif opt == '-i':
+            input_file = arg
+        elif opt == '-o':
+            output_file = arg
+        elif opt == '-d':
+            duration = arg
+
+    unwrapped_img = unwrap_img(input_file)
+    cv2.imwrite('output.png', unwrapped_img)
+
+    image_to_audio(unwrapped_img, output_file, duration, 44100, 20, 0.1)