Add Text Processor

Paragraph Identifier

TextProcessor.py

@@ -0,0 +1,83 @@
+import pandas as pd
+import numpy as np
+import fitz
+import re
+import cv2
+
+def paragraphs(folderpath):
+    doc = fitz.open(folderpath + '/opinion.pdf')
+    df = pd.read_csv(folderpath + '/data.csv').replace({np.nan: None})
+    indices = list(df.index)
+    pg_indices = df['Pg Ind'].tolist()
+
+    x1s, y1s, x2s, y2s, line_texts, line_inds, pg_inds, baselines = [], [], [], [], [], [], [], {}
+    paras = []
+    for (i, pg_ind) in enumerate(pg_indices):
+        lines = eval(df[df['Pg Ind'] == i]['Lines'].tolist()[0])
+        pg_x1s = []
+        for (j,n) in enumerate(lines):
+            x1s.append(n[0])
+            y1s.append(n[1])
+            x2s.append(n[2])
+            y2s.append(n[3])
+            line_texts.append(n[4])
+            pg_x1s.append(n[0])
+            pg_inds.append(i)
+            line_inds.append(j)
+        baselines[i] = min(pg_x1s)
+
+    for (j, line_text) in enumerate(line_texts):
+        if j == 0:
+            para = []
+            continue
+
+        if len(line_texts[j]) > 0:
+            prior_endswith_period = re.search('\.([^A-z]{0,2})$', line_texts[j-1].strip()) is not None
+
+            prior_is_section_header = re.search('^([ABCDEIVX]+)$', line_texts[j-1].strip()) is not None
+            current_is_section_header = re.search('^([ABCDEIVX]+)$', line_texts[j].strip()) is not None
+            prior_is_asterisk = re.search('^([\s\*]+)$', line_texts[j-1].strip()) is not None
+            current_is_asterisk = re.search('^([\s\*]+)$', line_texts[j].strip()) is not None
+            prior_is_date = re.search('(\[[A-z\s0-9]*,\s[0-9]*]+)$', line_texts[j-1].strip()) is not None
+
+            current_upper = line_text[0].isupper()
+            current_tabbed = x1s[j]-baselines[pg_inds[j]] > 7
+            prior_tabbed = x1s[j-1]-baselines[pg_inds[j-1]] > 7
+            prior_supertabbed = x1s[j-1]-baselines[pg_inds[j-1]] > 18
+            current_supertabbed = x1s[j]-baselines[pg_inds[j]] > 18
+            prior_more_left = x1s[j]-x1s[j-1] > 7
+
+            is_section_header = (prior_is_section_header or current_is_section_header or prior_is_asterisk or current_is_asterisk or prior_is_date)
+            prior_period_current_tabbed = (prior_endswith_period and current_tabbed and (prior_more_left or (not prior_supertabbed and not current_supertabbed)))
+
+            if is_section_header or prior_period_current_tabbed:
+                paras.append(para)
+                para = []
+            #     print('\n')
+            # print(str(j) + ':\t' + str(prior_endswith_period) + '\t' + str(current_tabbed)  + '\t' + str(prior_more_left)+ '\t' + str(prior_supertabbed) + '\t' + str(current_supertabbed) +'\t' + line_text)
+            para.append((pg_inds[j], line_inds[j], line_text))
+    paras.append(para)
+    paras_df = pd.DataFrame({'Lines': paras})
+    return paras_df
+
+def process_file(folderpath):
+    paras_df = paragraphs(folderpath)
+    paras_df.to_csv(folderpath + '/paragraphs.csv', index=True)
+    data_df = pd.read_csv(folderpath + '/data.csv')
+    paras_lines = paras_df['Lines'].tolist()
+    indents = []
+    for (i, para_lines) in enumerate(paras_lines):
+        para = []
+        para_start_pg_ind, para_start_line_ind, para_first_line = para_lines[0]
+        page_df = data_df[data_df['Pg Ind'] == para_start_pg_ind]
+        pg_lines = eval(page_df['Lines'].tolist()[0])
+        x1, y1, x2, y2, text = pg_lines[para_start_line_ind]
+        indents.append((x1, y1, x2, y2, para_first_line, para_start_pg_ind))
+
+    for indent in indents:
+        x1, y1, x2, y2, para_first_line, pg_ind = indent
+        image = cv2.imread(folderpath + '/' + str(pg_ind) + '-processed.png')
+        cv2.circle(image, (x1 - 15, int(0.5 * (y1 + y2))), radius=1, color=(240, 32, 160), thickness=2)
+        cv2.imwrite(folderpath + '/' + str(pg_ind) + '-processed.png', image)
+
+process_file('PDF Cases/333_178')