Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,7 +4,7 @@ import unicodedata
|
|
| 4 |
from paddleocr import PaddleOCR
|
| 5 |
|
| 6 |
# -------------------------------------------------
|
| 7 |
-
# OCR
|
| 8 |
# -------------------------------------------------
|
| 9 |
ocr = PaddleOCR(
|
| 10 |
lang="fr",
|
|
@@ -12,7 +12,7 @@ ocr = PaddleOCR(
|
|
| 12 |
)
|
| 13 |
|
| 14 |
# -------------------------------------------------
|
| 15 |
-
# Normalisation texte
|
| 16 |
# -------------------------------------------------
|
| 17 |
def normalize(text: str) -> str:
|
| 18 |
text = text.lower()
|
|
@@ -21,33 +21,19 @@ def normalize(text: str) -> str:
|
|
| 21 |
return " ".join(text.split())
|
| 22 |
|
| 23 |
# -------------------------------------------------
|
| 24 |
-
#
|
| 25 |
-
# -------------------------------------------------
|
| 26 |
-
COL_TITLES = {
|
| 27 |
-
"designation",
|
| 28 |
-
"designations",
|
| 29 |
-
"description",
|
| 30 |
-
"description des services"
|
| 31 |
-
}
|
| 32 |
-
|
| 33 |
-
# -------------------------------------------------
|
| 34 |
-
# Mots / lignes à ignorer
|
| 35 |
# -------------------------------------------------
|
|
|
|
| 36 |
IGNORE_KEYWORDS = {
|
| 37 |
-
"prix", "total", "ht", "htva", "tva",
|
| 38 |
-
"
|
| 39 |
}
|
| 40 |
|
| 41 |
-
|
| 42 |
-
#
|
| 43 |
-
# -------------------------------------------------
|
| 44 |
-
META_KEYWORDS = {
|
| 45 |
-
"dpo", "dao", "ref", "reference",
|
| 46 |
-
"date", "nme", ":"
|
| 47 |
-
}
|
| 48 |
|
| 49 |
# -------------------------------------------------
|
| 50 |
-
#
|
| 51 |
# -------------------------------------------------
|
| 52 |
def extract_second_column(image):
|
| 53 |
if image is None:
|
|
@@ -60,8 +46,8 @@ def extract_second_column(image):
|
|
| 60 |
return "OCR : aucun texte détecté."
|
| 61 |
|
| 62 |
data = result[0]
|
| 63 |
-
texts = data
|
| 64 |
-
boxes = data
|
| 65 |
|
| 66 |
blocks = []
|
| 67 |
for text, box in zip(texts, boxes):
|
|
@@ -74,61 +60,52 @@ def extract_second_column(image):
|
|
| 74 |
|
| 75 |
blocks.append((t, x, y))
|
| 76 |
|
| 77 |
-
if len(blocks) < 5:
|
| 78 |
-
return "Pas assez de texte exploitable."
|
| 79 |
-
|
| 80 |
# -------------------------------------------------
|
| 81 |
-
# 1.
|
| 82 |
# -------------------------------------------------
|
| 83 |
-
col_x = None
|
| 84 |
title_y = None
|
|
|
|
| 85 |
|
| 86 |
for text, x, y in blocks:
|
| 87 |
-
if normalize(text)
|
| 88 |
col_x = x
|
| 89 |
title_y = y
|
| 90 |
break
|
| 91 |
|
| 92 |
if col_x is None:
|
| 93 |
-
return "Titre
|
| 94 |
|
| 95 |
# -------------------------------------------------
|
| 96 |
-
# 2.
|
| 97 |
# -------------------------------------------------
|
| 98 |
-
X_THRESHOLD = 45
|
| 99 |
column_blocks = [
|
| 100 |
(t, x, y) for t, x, y in blocks
|
| 101 |
-
if abs(x - col_x) < X_THRESHOLD and y > title_y
|
| 102 |
]
|
| 103 |
|
| 104 |
if not column_blocks:
|
| 105 |
-
return "
|
| 106 |
|
| 107 |
# -------------------------------------------------
|
| 108 |
-
# 3. Tri vertical
|
| 109 |
# -------------------------------------------------
|
| 110 |
column_blocks.sort(key=lambda e: e[2])
|
| 111 |
|
| 112 |
# -------------------------------------------------
|
| 113 |
-
# 4. Fusion contrôlée
|
| 114 |
# -------------------------------------------------
|
| 115 |
merged = []
|
| 116 |
current = ""
|
| 117 |
last_y = None
|
| 118 |
-
Y_THRESHOLD = 22
|
| 119 |
|
| 120 |
for text, x, y in column_blocks:
|
| 121 |
nt = normalize(text)
|
| 122 |
|
| 123 |
-
# Ignore lignes de totaux / prix
|
| 124 |
if any(k in nt for k in IGNORE_KEYWORDS):
|
| 125 |
continue
|
| 126 |
|
| 127 |
-
#
|
| 128 |
-
if
|
| 129 |
-
continue
|
| 130 |
-
|
| 131 |
-
if last_y is None or abs(y - last_y) > Y_THRESHOLD:
|
| 132 |
if current:
|
| 133 |
merged.append(current.strip())
|
| 134 |
current = text
|
|
@@ -141,40 +118,33 @@ def extract_second_column(image):
|
|
| 141 |
merged.append(current.strip())
|
| 142 |
|
| 143 |
# -------------------------------------------------
|
| 144 |
-
# 5. Nettoyage final
|
| 145 |
# -------------------------------------------------
|
| 146 |
final = []
|
| 147 |
for line in merged:
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
if len(nt) < 4:
|
| 151 |
continue
|
| 152 |
|
| 153 |
-
if sum(c.isdigit() for c in line) > len(line)
|
| 154 |
continue
|
| 155 |
|
| 156 |
final.append(line)
|
| 157 |
|
| 158 |
if not final:
|
| 159 |
-
return "Aucune cellule
|
| 160 |
|
| 161 |
-
# -------------------------------------------------
|
| 162 |
-
# 6. Résultat numéroté
|
| 163 |
-
# -------------------------------------------------
|
| 164 |
return "\n".join(f"{i+1}. {line}" for i, line in enumerate(final))
|
| 165 |
|
|
|
|
| 166 |
# -------------------------------------------------
|
| 167 |
-
# Interface Gradio
|
| 168 |
# -------------------------------------------------
|
| 169 |
demo = gr.Interface(
|
| 170 |
fn=extract_second_column,
|
| 171 |
inputs=gr.Image(type="pil", label="Image du tableau"),
|
| 172 |
outputs=gr.Textbox(label="Contenu de la colonne 2"),
|
| 173 |
-
title="Extraction fiable
|
| 174 |
-
description=
|
| 175 |
-
"Extraction robuste de la deuxième colonne des tableaux scannés "
|
| 176 |
-
"(Désignation, DESIGNATIONS, Description, Description des services)."
|
| 177 |
-
)
|
| 178 |
)
|
| 179 |
|
| 180 |
demo.launch(server_name="0.0.0.0", server_port=7860)
|
|
|
|
| 4 |
from paddleocr import PaddleOCR
|
| 5 |
|
| 6 |
# -------------------------------------------------
|
| 7 |
+
# OCR
|
| 8 |
# -------------------------------------------------
|
| 9 |
ocr = PaddleOCR(
|
| 10 |
lang="fr",
|
|
|
|
| 12 |
)
|
| 13 |
|
| 14 |
# -------------------------------------------------
|
| 15 |
+
# Normalisation texte
|
| 16 |
# -------------------------------------------------
|
| 17 |
def normalize(text: str) -> str:
|
| 18 |
text = text.lower()
|
|
|
|
| 21 |
return " ".join(text.split())
|
| 22 |
|
| 23 |
# -------------------------------------------------
|
| 24 |
+
# Paramètres spécifiques image "Description des services"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
# -------------------------------------------------
|
| 26 |
+
COLUMN_TITLE = "description des services"
|
| 27 |
IGNORE_KEYWORDS = {
|
| 28 |
+
"prix", "total", "ht", "htva", "tva", "ttc",
|
| 29 |
+
"general", "generale"
|
| 30 |
}
|
| 31 |
|
| 32 |
+
X_THRESHOLD = 45
|
| 33 |
+
Y_NEW_CELL = 32 # seuil volontairement élevé → empêche fusion abusive
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
|
| 35 |
# -------------------------------------------------
|
| 36 |
+
# Extraction colonne 2
|
| 37 |
# -------------------------------------------------
|
| 38 |
def extract_second_column(image):
|
| 39 |
if image is None:
|
|
|
|
| 46 |
return "OCR : aucun texte détecté."
|
| 47 |
|
| 48 |
data = result[0]
|
| 49 |
+
texts = data["rec_texts"]
|
| 50 |
+
boxes = data["dt_polys"]
|
| 51 |
|
| 52 |
blocks = []
|
| 53 |
for text, box in zip(texts, boxes):
|
|
|
|
| 60 |
|
| 61 |
blocks.append((t, x, y))
|
| 62 |
|
|
|
|
|
|
|
|
|
|
| 63 |
# -------------------------------------------------
|
| 64 |
+
# 1. Trouver le titre exact de la colonne
|
| 65 |
# -------------------------------------------------
|
|
|
|
| 66 |
title_y = None
|
| 67 |
+
col_x = None
|
| 68 |
|
| 69 |
for text, x, y in blocks:
|
| 70 |
+
if normalize(text) == COLUMN_TITLE:
|
| 71 |
col_x = x
|
| 72 |
title_y = y
|
| 73 |
break
|
| 74 |
|
| 75 |
if col_x is None:
|
| 76 |
+
return "Titre 'Description des services' non détecté."
|
| 77 |
|
| 78 |
# -------------------------------------------------
|
| 79 |
+
# 2. Garder uniquement le texte SOUS le titre
|
| 80 |
# -------------------------------------------------
|
|
|
|
| 81 |
column_blocks = [
|
| 82 |
(t, x, y) for t, x, y in blocks
|
| 83 |
+
if abs(x - col_x) < X_THRESHOLD and y > title_y + 15
|
| 84 |
]
|
| 85 |
|
| 86 |
if not column_blocks:
|
| 87 |
+
return "Aucune cellule détectée sous la colonne."
|
| 88 |
|
| 89 |
# -------------------------------------------------
|
| 90 |
+
# 3. Tri vertical
|
| 91 |
# -------------------------------------------------
|
| 92 |
column_blocks.sort(key=lambda e: e[2])
|
| 93 |
|
| 94 |
# -------------------------------------------------
|
| 95 |
+
# 4. Fusion contrôlée (évite fusion Tuyau / Coude)
|
| 96 |
# -------------------------------------------------
|
| 97 |
merged = []
|
| 98 |
current = ""
|
| 99 |
last_y = None
|
|
|
|
| 100 |
|
| 101 |
for text, x, y in column_blocks:
|
| 102 |
nt = normalize(text)
|
| 103 |
|
|
|
|
| 104 |
if any(k in nt for k in IGNORE_KEYWORDS):
|
| 105 |
continue
|
| 106 |
|
| 107 |
+
# nouvelle cellule
|
| 108 |
+
if last_y is None or abs(y - last_y) > Y_NEW_CELL:
|
|
|
|
|
|
|
|
|
|
| 109 |
if current:
|
| 110 |
merged.append(current.strip())
|
| 111 |
current = text
|
|
|
|
| 118 |
merged.append(current.strip())
|
| 119 |
|
| 120 |
# -------------------------------------------------
|
| 121 |
+
# 5. Nettoyage final
|
| 122 |
# -------------------------------------------------
|
| 123 |
final = []
|
| 124 |
for line in merged:
|
| 125 |
+
if not line[0].isupper():
|
|
|
|
|
|
|
| 126 |
continue
|
| 127 |
|
| 128 |
+
if sum(c.isdigit() for c in line) > len(line) * 0.3:
|
| 129 |
continue
|
| 130 |
|
| 131 |
final.append(line)
|
| 132 |
|
| 133 |
if not final:
|
| 134 |
+
return "Aucune cellule valide trouvée."
|
| 135 |
|
|
|
|
|
|
|
|
|
|
| 136 |
return "\n".join(f"{i+1}. {line}" for i, line in enumerate(final))
|
| 137 |
|
| 138 |
+
|
| 139 |
# -------------------------------------------------
|
| 140 |
+
# Interface Gradio
|
| 141 |
# -------------------------------------------------
|
| 142 |
demo = gr.Interface(
|
| 143 |
fn=extract_second_column,
|
| 144 |
inputs=gr.Image(type="pil", label="Image du tableau"),
|
| 145 |
outputs=gr.Textbox(label="Contenu de la colonne 2"),
|
| 146 |
+
title="Extraction fiable – Colonne 2 (Description des services)",
|
| 147 |
+
description="Extraction robuste et ordonnée des cellules texte."
|
|
|
|
|
|
|
|
|
|
| 148 |
)
|
| 149 |
|
| 150 |
demo.launch(server_name="0.0.0.0", server_port=7860)
|