Color FIX

app.py

@@ -3,67 +3,49 @@ import cv2, numpy as np
 from PIL import Image
 
 # -------------------------------------------------
-# 🔧 CORE FUNCTION
+# Core Fold Removal Function
 # -------------------------------------------------
-def remove_folds(image, intensity=0.5):
+def remove_folds_color_preserve(image, intensity=0.5, enhance=False):
     if image is None:
         return None
 
-    img = np.array(image.convert("RGB"))
-    img_lab = cv2.cvtColor(img, cv2.COLOR_RGB2LAB)
-    l, a, b = cv2.split(img_lab)
+    img = np.array(image.convert("RGB"), dtype=np.float32) / 255.0
 
-    # --- (1) Smooth illumination correction ---
-    ksize = int(49 + intensity * 100)
-    ksize = ksize + 1 if ksize % 2 == 0 else ksize
-    l_blur = cv2.GaussianBlur(l, (ksize, ksize), 0)
-    l_equal = cv2.divide(l, l_blur, scale=128)
+    # --- (1) Estimate illumination via large guided blur ---
+    radius = int(80 + 120 * intensity)
+    illum = cv2.GaussianBlur(img, (radius|1, radius|1), 0)
 
-    # --- (2) Frequency-domain smoothing ---
-    freq = np.fft.fft2(l_equal)
-    freqshift = np.fft.fftshift(freq)
-    rows, cols = l_equal.shape
-    crow, ccol = rows // 2, cols // 2
-    mask = np.ones((rows, cols), np.uint8)
-    r = int(10 + intensity * 30)
-    mask[crow - r:crow + r, ccol - r:ccol + r] = 0
-    freqshift = freqshift * mask
-    ishift = np.fft.ifftshift(freqshift)
-    l_flat = np.abs(np.fft.ifft2(ishift))
-    l_flat = cv2.normalize(l_flat, None, 0, 255, cv2.NORM_MINMAX).astype(np.uint8)
+    # --- (2) Divide reflectance by illumination (Retinex-like) ---
+    norm = np.clip(img / (illum + 1e-5), 0, 4)
+    norm = norm / norm.max()
 
-    # --- (3) Merge back and convert ---
-    img_lab = cv2.merge([l_flat, a, b])
-    result = cv2.cvtColor(img_lab, cv2.COLOR_LAB2RGB)
-    return Image.fromarray(result)
+    # --- (3) Edge-aware enhancement to restore motifs ---
+    out = cv2.detailEnhance((norm * 255).astype(np.uint8), sigma_s=20, sigma_r=0.2)
+
+    # --- (4) Optional weave enhancement ---
+    if enhance:
+        blur = cv2.GaussianBlur(out, (0, 0), 3)
+        out = cv2.addWeighted(out, 1.5, blur, -0.5, 0)
+
+    return Image.fromarray(out)
 
 # -------------------------------------------------
-# 🎛️ GRADIO BLOCKS UI (with slider throttle)
+# Gradio Blocks UI
 # -------------------------------------------------
-with gr.Blocks(title="🪄 Saree Fold Remover (Lighting Only, No Distortion)") as demo:
-    gr.Markdown(
-        """
-        ## 🪄 Saree Fold Remover  
-        Upload a flat saree or fabric image.<br>
-        Adjust the **Fold Intensity** slider — higher values remove deeper folds.<br>
-        Lighting only correction, no geometric distortion.
-        """
-    )
+with gr.Blocks(title="🪄 Saree Fold Remover — Color-Preserving Edition") as demo:
+    gr.Markdown("""
+    ## 🪄 Saree Fold Remover (Color-Preserving)
+    Removes lighting folds & shadows **without color shifts or distortion**.<br>
+    Use the slider to control strength, and enable *Enhance Texture* for fine weave sharpening.
+    """)
 
     with gr.Row():
         inp = gr.Image(label="Upload Saree Image", type="pil")
-        intensity = gr.Slider(
-            0.0, 1.0, value=0.5, step=0.05,
-            label="Fold Intensity",
-            info="Higher = stronger flattening (lighting only)"
-        )
+        intensity = gr.Slider(0.0, 1.0, value=0.5, step=0.05, label="Fold Intensity")
+        enhance = gr.Checkbox(label="✨ Enhance Texture", value=False)
 
     out = gr.Image(label="Flat, Fold-Free Output")
-
-    # Button to trigger with throttle behavior
-    run_btn = gr.Button("✨ Remove Folds")
-
-    # Run when button is pressed (instead of live refresh)
-    run_btn.click(fn=remove_folds, inputs=[inp, intensity], outputs=out)
+    run = gr.Button("🚀 Remove Folds")
+    run.click(fn=remove_folds_color_preserve, inputs=[inp, intensity, enhance], outputs=out)
 
 demo.launch()