Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -3,26 +3,26 @@ import numpy as np
|
|
| 3 |
import unicodedata
|
| 4 |
from paddleocr import PaddleOCR
|
| 5 |
|
| 6 |
-
#
|
| 7 |
-
# OCR
|
| 8 |
-
#
|
| 9 |
ocr = PaddleOCR(
|
| 10 |
lang="fr",
|
| 11 |
use_textline_orientation=True
|
| 12 |
)
|
| 13 |
|
| 14 |
-
#
|
| 15 |
# Normalisation texte (casse + accents)
|
| 16 |
-
#
|
| 17 |
def normalize(text: str) -> str:
|
| 18 |
text = text.lower()
|
| 19 |
text = unicodedata.normalize("NFD", text)
|
| 20 |
text = "".join(c for c in text if unicodedata.category(c) != "Mn")
|
| 21 |
return " ".join(text.split())
|
| 22 |
|
| 23 |
-
#
|
| 24 |
-
# Titres
|
| 25 |
-
#
|
| 26 |
COL_TITLES = {
|
| 27 |
"designation",
|
| 28 |
"designations",
|
|
@@ -30,34 +30,32 @@ COL_TITLES = {
|
|
| 30 |
"description des services"
|
| 31 |
}
|
| 32 |
|
| 33 |
-
#
|
| 34 |
-
# Mots à ignorer
|
| 35 |
-
#
|
| 36 |
IGNORE_KEYWORDS = {
|
| 37 |
-
"prix", "
|
| 38 |
-
"
|
| 39 |
}
|
| 40 |
|
| 41 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
# Fonction principale
|
| 43 |
-
#
|
| 44 |
def extract_second_column(image):
|
| 45 |
if image is None:
|
| 46 |
return "Aucune image fournie."
|
| 47 |
|
| 48 |
img = np.array(image)
|
| 49 |
-
|
| 50 |
-
# -------------------------------------------------
|
| 51 |
-
# Rotation automatique si image couchée
|
| 52 |
-
# -------------------------------------------------
|
| 53 |
-
h, w = img.shape[:2]
|
| 54 |
-
if w > h:
|
| 55 |
-
img = np.rot90(img, 1)
|
| 56 |
-
|
| 57 |
-
# -------------------------------------------------
|
| 58 |
-
# OCR
|
| 59 |
-
# -------------------------------------------------
|
| 60 |
result = ocr.predict(img)
|
|
|
|
| 61 |
if not result:
|
| 62 |
return "OCR : aucun texte détecté."
|
| 63 |
|
|
@@ -76,56 +74,60 @@ def extract_second_column(image):
|
|
| 76 |
|
| 77 |
blocks.append((t, x, y))
|
| 78 |
|
| 79 |
-
if len(blocks) <
|
| 80 |
return "Pas assez de texte exploitable."
|
| 81 |
|
| 82 |
# -------------------------------------------------
|
| 83 |
-
# Détection du X de la colonne
|
| 84 |
# -------------------------------------------------
|
| 85 |
col_x = None
|
|
|
|
|
|
|
| 86 |
for text, x, y in blocks:
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
col_x = x
|
| 91 |
-
break
|
| 92 |
-
if col_x is not None:
|
| 93 |
break
|
| 94 |
|
| 95 |
if col_x is None:
|
| 96 |
return "Titre de la colonne cible non détecté."
|
| 97 |
|
| 98 |
# -------------------------------------------------
|
| 99 |
-
# Sélection des blocs de la colonne
|
| 100 |
# -------------------------------------------------
|
| 101 |
-
X_THRESHOLD =
|
| 102 |
column_blocks = [
|
| 103 |
(t, x, y) for t, x, y in blocks
|
| 104 |
-
if abs(x - col_x) < X_THRESHOLD
|
| 105 |
]
|
| 106 |
|
| 107 |
if not column_blocks:
|
| 108 |
return "Colonne détectée mais vide."
|
| 109 |
|
| 110 |
# -------------------------------------------------
|
| 111 |
-
# Tri vertical
|
| 112 |
# -------------------------------------------------
|
| 113 |
column_blocks.sort(key=lambda e: e[2])
|
| 114 |
|
| 115 |
# -------------------------------------------------
|
| 116 |
-
# Fusion
|
| 117 |
# -------------------------------------------------
|
| 118 |
merged = []
|
| 119 |
current = ""
|
| 120 |
last_y = None
|
| 121 |
-
Y_THRESHOLD =
|
| 122 |
|
| 123 |
for text, x, y in column_blocks:
|
| 124 |
nt = normalize(text)
|
| 125 |
|
|
|
|
| 126 |
if any(k in nt for k in IGNORE_KEYWORDS):
|
| 127 |
continue
|
| 128 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 129 |
if last_y is None or abs(y - last_y) > Y_THRESHOLD:
|
| 130 |
if current:
|
| 131 |
merged.append(current.strip())
|
|
@@ -139,19 +141,16 @@ def extract_second_column(image):
|
|
| 139 |
merged.append(current.strip())
|
| 140 |
|
| 141 |
# -------------------------------------------------
|
| 142 |
-
# Nettoyage final
|
| 143 |
# -------------------------------------------------
|
| 144 |
final = []
|
| 145 |
for line in merged:
|
| 146 |
nt = normalize(line)
|
| 147 |
|
| 148 |
-
if nt
|
| 149 |
-
continue
|
| 150 |
-
if len(nt) < 5:
|
| 151 |
continue
|
| 152 |
-
|
| 153 |
-
|
| 154 |
-
if not line[0].isupper():
|
| 155 |
continue
|
| 156 |
|
| 157 |
final.append(line)
|
|
@@ -159,18 +158,21 @@ def extract_second_column(image):
|
|
| 159 |
if not final:
|
| 160 |
return "Aucune cellule texte valide trouvée."
|
| 161 |
|
|
|
|
|
|
|
|
|
|
| 162 |
return "\n".join(f"{i+1}. {line}" for i, line in enumerate(final))
|
| 163 |
|
| 164 |
-
#
|
| 165 |
-
# Interface Gradio
|
| 166 |
-
#
|
| 167 |
demo = gr.Interface(
|
| 168 |
fn=extract_second_column,
|
| 169 |
inputs=gr.Image(type="pil", label="Image du tableau"),
|
| 170 |
outputs=gr.Textbox(label="Contenu de la colonne 2"),
|
| 171 |
title="Extraction fiable de la colonne 2",
|
| 172 |
description=(
|
| 173 |
-
"Extraction
|
| 174 |
"(Désignation, DESIGNATIONS, Description, Description des services)."
|
| 175 |
)
|
| 176 |
)
|
|
|
|
| 3 |
import unicodedata
|
| 4 |
from paddleocr import PaddleOCR
|
| 5 |
|
| 6 |
+
# -------------------------------------------------
|
| 7 |
+
# OCR (compatible Hugging Face)
|
| 8 |
+
# -------------------------------------------------
|
| 9 |
ocr = PaddleOCR(
|
| 10 |
lang="fr",
|
| 11 |
use_textline_orientation=True
|
| 12 |
)
|
| 13 |
|
| 14 |
+
# -------------------------------------------------
|
| 15 |
# Normalisation texte (casse + accents)
|
| 16 |
+
# -------------------------------------------------
|
| 17 |
def normalize(text: str) -> str:
|
| 18 |
text = text.lower()
|
| 19 |
text = unicodedata.normalize("NFD", text)
|
| 20 |
text = "".join(c for c in text if unicodedata.category(c) != "Mn")
|
| 21 |
return " ".join(text.split())
|
| 22 |
|
| 23 |
+
# -------------------------------------------------
|
| 24 |
+
# Titres valides de la colonne 2
|
| 25 |
+
# -------------------------------------------------
|
| 26 |
COL_TITLES = {
|
| 27 |
"designation",
|
| 28 |
"designations",
|
|
|
|
| 30 |
"description des services"
|
| 31 |
}
|
| 32 |
|
| 33 |
+
# -------------------------------------------------
|
| 34 |
+
# Mots / lignes à ignorer
|
| 35 |
+
# -------------------------------------------------
|
| 36 |
IGNORE_KEYWORDS = {
|
| 37 |
+
"prix", "total", "ht", "htva", "tva",
|
| 38 |
+
"ttc", "general", "generale"
|
| 39 |
}
|
| 40 |
|
| 41 |
+
# -------------------------------------------------
|
| 42 |
+
# Métadonnées à exclure (hors tableau)
|
| 43 |
+
# -------------------------------------------------
|
| 44 |
+
META_KEYWORDS = {
|
| 45 |
+
"dpo", "dao", "ref", "reference",
|
| 46 |
+
"date", "nme", ":"
|
| 47 |
+
}
|
| 48 |
+
|
| 49 |
+
# -------------------------------------------------
|
| 50 |
# Fonction principale
|
| 51 |
+
# -------------------------------------------------
|
| 52 |
def extract_second_column(image):
|
| 53 |
if image is None:
|
| 54 |
return "Aucune image fournie."
|
| 55 |
|
| 56 |
img = np.array(image)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
result = ocr.predict(img)
|
| 58 |
+
|
| 59 |
if not result:
|
| 60 |
return "OCR : aucun texte détecté."
|
| 61 |
|
|
|
|
| 74 |
|
| 75 |
blocks.append((t, x, y))
|
| 76 |
|
| 77 |
+
if len(blocks) < 5:
|
| 78 |
return "Pas assez de texte exploitable."
|
| 79 |
|
| 80 |
# -------------------------------------------------
|
| 81 |
+
# 1. Détection du X de la colonne cible (par le titre)
|
| 82 |
# -------------------------------------------------
|
| 83 |
col_x = None
|
| 84 |
+
title_y = None
|
| 85 |
+
|
| 86 |
for text, x, y in blocks:
|
| 87 |
+
if normalize(text) in COL_TITLES:
|
| 88 |
+
col_x = x
|
| 89 |
+
title_y = y
|
|
|
|
|
|
|
|
|
|
| 90 |
break
|
| 91 |
|
| 92 |
if col_x is None:
|
| 93 |
return "Titre de la colonne cible non détecté."
|
| 94 |
|
| 95 |
# -------------------------------------------------
|
| 96 |
+
# 2. Sélection des blocs de la colonne (SOUS le titre)
|
| 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 "Colonne détectée mais vide."
|
| 106 |
|
| 107 |
# -------------------------------------------------
|
| 108 |
+
# 3. Tri vertical (haut → bas)
|
| 109 |
# -------------------------------------------------
|
| 110 |
column_blocks.sort(key=lambda e: e[2])
|
| 111 |
|
| 112 |
# -------------------------------------------------
|
| 113 |
+
# 4. Fusion contrôlée des lignes OCR
|
| 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 |
+
# Ignore métadonnées résiduelles
|
| 128 |
+
if any(k in nt for k in META_KEYWORDS):
|
| 129 |
+
continue
|
| 130 |
+
|
| 131 |
if last_y is None or abs(y - last_y) > Y_THRESHOLD:
|
| 132 |
if current:
|
| 133 |
merged.append(current.strip())
|
|
|
|
| 141 |
merged.append(current.strip())
|
| 142 |
|
| 143 |
# -------------------------------------------------
|
| 144 |
+
# 5. Nettoyage final (cellules texte métier uniquement)
|
| 145 |
# -------------------------------------------------
|
| 146 |
final = []
|
| 147 |
for line in merged:
|
| 148 |
nt = normalize(line)
|
| 149 |
|
| 150 |
+
if len(nt) < 4:
|
|
|
|
|
|
|
| 151 |
continue
|
| 152 |
+
|
| 153 |
+
if sum(c.isdigit() for c in line) > len(line) / 2:
|
|
|
|
| 154 |
continue
|
| 155 |
|
| 156 |
final.append(line)
|
|
|
|
| 158 |
if not final:
|
| 159 |
return "Aucune cellule texte valide trouvée."
|
| 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 (Hugging Face)
|
| 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 de la colonne 2",
|
| 174 |
description=(
|
| 175 |
+
"Extraction robuste de la deuxième colonne des tableaux scannés "
|
| 176 |
"(Désignation, DESIGNATIONS, Description, Description des services)."
|
| 177 |
)
|
| 178 |
)
|