Fix bounding box offsets, scale clamping, and column-aware reading order

gradio_api.py

@@ -398,32 +398,38 @@ def insert_ocr_text_box(page, font, text, p0, p1, p3, zoom=1.0):
     ascender = font.ascender if hasattr(font, 'ascender') and font.ascender > 0 else 0.928
     descender = font.descender if hasattr(font, 'descender') and font.descender < 0 else -0.236
     font_ratio = ascender - descender
-    
-    fontsize = max(min(box_height_pdf / font_ratio, 20.0), 5.5)
+    fontsize = max(min(box_height_pdf / font_ratio, 13.5), 5.5)
     
     dx_up = p0_pdf[0] - p3_pdf[0]
     dy_up = p0_pdf[1] - p3_pdf[1]
     base_x = p3_pdf[0] + dx_up * (-descender / font_ratio)
     base_y = p3_pdf[1] + dy_up * (-descender / font_ratio)
     
-    base_x = max(page.rect.x0 + 5.0, min(page.rect.x1 - 15.0, base_x))
-    base_y = max(page.rect.y0 + 5.0, min(page.rect.y1 - 5.0, base_y))
-
-    text_len = font.text_length(text, fontsize=fontsize)
-    if text_len > 0 and box_width_pdf > 0:
-        raw_scale = box_width_pdf / text_len
-        scale_x = max(0.70, min(1.15, raw_scale))
+    max_right = page.rect.width - 35.0
+    max_bottom = page.rect.height - 25.0
+    base_x = max(page.rect.x0 + 15.0, min(max_right - 20.0, base_x))
+    base_y = max(page.rect.y0 + 15.0, min(max_bottom - 5.0, base_y))
+
+    natural_len = font.text_length(text, fontsize=fontsize)
+    if natural_len > 0 and box_width_pdf > 0:
+        raw_scale = box_width_pdf / natural_len
+        if 0.80 <= raw_scale <= 1.05:
+            scale_x = raw_scale
+        elif raw_scale > 1.05:
+            scale_x = 1.0
+        else:
+            scale_x = max(0.70, raw_scale)
     else:
         scale_x = 1.0
 
-    # STRICT RIGHT PAGE BOUNDARY CLAMP: Never allow text box to project into black background margin
-    max_allowed_w = page.rect.x1 - 5.0 - base_x
-    if max_allowed_w > 5.0 and (text_len * scale_x) > max_allowed_w:
-        scale_x = max_allowed_w / text_len if text_len > 0 else 1.0
-        if scale_x < 0.65:
-            fontsize = max(5.0, fontsize * (scale_x / 0.65))
-            text_len = font.text_length(text, fontsize=fontsize)
-            scale_x = max_allowed_w / text_len if text_len > 0 else 1.0
+    # STRICT RIGHT PAGE BOUNDARY CLAMP: Ensure (base_x + natural_len * scale_x) <= max_right
+    max_allowed_w = max(5.0, max_right - base_x)
+    if (natural_len * scale_x) > max_allowed_w:
+        scale_x = max_allowed_w / natural_len if natural_len > 0 else 1.0
+        if scale_x < 0.60:
+            fontsize = max(4.5, fontsize * (scale_x / 0.60))
+            natural_len = font.text_length(text, fontsize=fontsize)
+            scale_x = max_allowed_w / natural_len if natural_len > 0 else 1.0
 
     matrix = fitz.Matrix(scale_x, 1.0) * fitz.Matrix(-angle_deg)
     try:
@@ -444,25 +450,37 @@ def insert_clamped_rect_text(page, font, text, rect):
     """
     if not text or not text.strip():
         return
-    x0 = max(page.rect.x0 + 5.0, min(page.rect.x1 - 15.0, rect.x0))
-    y0 = max(page.rect.y0 + 5.0, min(page.rect.y1 - 5.0, rect.y0))
-    x1 = max(x0 + 10.0, min(page.rect.x1 - 5.0, rect.x1))
-    y1 = max(y0 + 5.0, min(page.rect.y1 - 5.0, rect.y1))
+    max_right = page.rect.width - 35.0
+    max_bottom = page.rect.height - 25.0
+
+    x0 = max(page.rect.x0 + 15.0, min(max_right - 20.0, rect.x0))
+    y0 = max(page.rect.y0 + 15.0, min(max_bottom - 10.0, rect.y0))
+    x1 = max(x0 + 10.0, min(max_right, rect.x1))
+    y1 = max(y0 + 5.0, min(max_bottom, rect.y1))
 
     box_w = x1 - x0
     box_h = y1 - y0
 
-    fontsize = max(min(box_h / 1.15, 20.0), 5.5)
-    text_len = font.text_length(text, fontsize=fontsize)
-    scale_x = (box_w / text_len) if text_len > 0 else 1.0
+    fontsize = max(min(box_h * 0.75, 13.5), 5.5)
+    natural_len = font.text_length(text, fontsize=fontsize)
+    if natural_len > 0 and box_w > 0:
+        raw_scale = box_w / natural_len
+        if 0.80 <= raw_scale <= 1.05:
+            scale_x = raw_scale
+        elif raw_scale > 1.05:
+            scale_x = 1.0
+        else:
+            scale_x = max(0.70, raw_scale)
+    else:
+        scale_x = 1.0
 
-    max_allowed_w = page.rect.x1 - 5.0 - x0
-    if max_allowed_w > 5.0 and (text_len * scale_x) > max_allowed_w:
-        scale_x = max_allowed_w / text_len if text_len > 0 else 1.0
-        if scale_x < 0.65:
-            fontsize = max(5.0, fontsize * (scale_x / 0.65))
-            text_len = font.text_length(text, fontsize=fontsize)
-            scale_x = max_allowed_w / text_len if text_len > 0 else 1.0
+    max_allowed_w = max(5.0, max_right - x0)
+    if (natural_len * scale_x) > max_allowed_w:
+        scale_x = max_allowed_w / natural_len if natural_len > 0 else 1.0
+        if scale_x < 0.60:
+            fontsize = max(4.5, fontsize * (scale_x / 0.60))
+            natural_len = font.text_length(text, fontsize=fontsize)
+            scale_x = max_allowed_w / natural_len if natural_len > 0 else 1.0
 
     y_baseline = y1 - (box_h * 0.18)
     point = fitz.Point(x0, y_baseline)
@@ -874,6 +892,10 @@ def process_pdf(input_file_path, mode, smart_skip=True, user_gemini_key=None, pr
                 min_y = min(p0_pdf[1], p1_pdf[1], p2_pdf[1], p3_pdf[1])
                 max_y = max(p0_pdf[1], p1_pdf[1], p2_pdf[1], p3_pdf[1])
                 
+                # Filter out scanner edge noise boxes outside page printable area
+                if min_x > (page.rect.width - 30.0) or max_x < 15.0 or min_y > (page.rect.height - 20.0):
+                    continue
+
                 pdf_rect = fitz.Rect(min_x, min_y, max_x, max_y)
                 pdf_paddle_items.append({
                     "poly": (p0, p1, p2, p3),
@@ -1040,11 +1062,16 @@ def process_pdf(input_file_path, mode, smart_skip=True, user_gemini_key=None, pr
                 text = line[1][0]
                 if not text.strip(): continue
                 
-                xmin = min(p[0] for p in box)
-                ymin = min(p[1] for p in box)
-                xmax = max(p[0] for p in box)
-                ymax = max(p[1] for p in box)
-                boxes_with_data.append(([xmin, ymin, xmax, ymax], (box, text)))
+                xmin = min(p[0] for p in box) / zoom
+                ymin = min(p[1] for p in box) / zoom
+                xmax = max(p[0] for p in box) / zoom
+                ymax = max(p[1] for p in box) / zoom
+
+                # Filter out scanner edge noise boxes outside page printable area
+                if xmin > (page.rect.width - 30.0) or xmax < 15.0 or ymin > (page.rect.height - 20.0):
+                    continue
+
+                boxes_with_data.append(([xmin * zoom, ymin * zoom, xmax * zoom, ymax * zoom], (box, text)))
                 
             sorted_data = recursive_xy_cut(boxes_with_data)