Improve text filtering with OpenCV

rok-reader.py

 import argparse
+import cv2
 import glob
 import os
 import multiprocessing
+import numpy
 from PIL import Image, ImageFilter
-from PIL.ImageOps import autocontrast, invert, grayscale, contain
+from PIL.ImageOps import autocontrast, invert, grayscale, contain, expand
 import pytesseract
 import signal
 import sys
 
+
 # ----
 # Program
 # ----
 
 KILLS_TARGETS = [
     ("Kill Points", (1418, 312), (1694, 352), True, False, 0),
-    ("T1 Kills", (1321, 637), (1538, 684), True, False, 0),
+    ("T1 Kills", (1325, 637), (1538, 684), True, False, 0),
     ("T1 Kill Points", (1986, 637), (2212, 684), True, False, 0),
-    ("T2 Kills", (1321, 702), (1538, 755), True, False, 0),
+    ("T2 Kills", (1325, 702), (1538, 755), True, False, 0),
     ("T2 Kill Points", (1986, 702), (2212, 755), True, False, 0),
-    ("T3 Kills", (1321, 770), (1538, 824), True, False, 0),
+    ("T3 Kills", (1325, 770), (1538, 824), True, False, 0),
     ("T3 Kill Points", (1986, 770), (2212, 824), True, False, 0),
-    ("T4 Kills", (1321, 847), (1538, 897), True, False, 0),
+    ("T4 Kills", (1325, 847), (1538, 897), True, False, 0),
     ("T4 Kill Points", (1986, 847), (2212, 897), True, False, 0),
-    ("T5 Kills", (1321, 918), (1538, 968), True, False, 0),
+    ("T5 Kills", (1325, 918), (1538, 968), True, False, 0),
     ("T5 Kill Points", (1986, 918), (2212, 968), True, False, 0),
-    ("Previous Kills", (1626, 985), (2228, 1039), False, False, 0)
+    ("Previous Kills", (1626, 985), (2228, 1039), False, False, -385)
 ]
 
 # ----
 
 # Read text from a section of an image using Tesseract
 def read_string_from_image(rgbImage, box, is_number, inv, bonusRightTrim, debugFilePath):
-        # Crop to correct dimentions
+        # Crop to correct dimensions
         rgbImage = rgbImage.crop((box.x, box.y, box.x2, box.y2))
 
         # Invert if flagged
         if inv: rgbImage = invert(rgbImage)
 
-        # Apply filters
-        rgbImage = grayscale(rgbImage)
-        rgbImage = autocontrast(rgbImage, cutoff=(0, 75))
+        # Apply Pillow filters to cut off artifacts
+        rgbImage = autocontrast(rgbImage, cutoff=(0, 50))
+
+        # Convert to OpenCV
+        npImage=numpy.array(rgbImage)
+
+        # Set colors to grayscale
+        npImage=cv2.cvtColor(npImage, cv2.COLOR_BGR2GRAY)
+
+        # Apply OpenCV Filters
+        npImage = cv2.medianBlur(npImage, 3)
+        _, npImage = cv2.threshold(npImage, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
+        # npImage = cv2.adaptiveThreshold(npImage, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
+
+        # Convert to Pillow
+        rgbImage = Image.fromarray(npImage)
 
         # Crop to content
-        bbox = autocontrast(invert(rgbImage), cutoff=(0, 90)).getbbox()
+        bbox = invert(rgbImage).getbbox()
         if bbox: rgbImage = rgbImage.crop((bbox[0], bbox[1], bbox[2] + bonusRightTrim, bbox[3]))
-
-        # Resize and sharpen
-        rgbImage = contain(rgbImage, (800, 800), method=1)
-        rgbImage = rgbImage.filter(ImageFilter.EDGE_ENHANCE_MORE)
-        rgbImage = rgbImage.filter(ImageFilter.SHARPEN)
+        if bbox: rgbImage = rgbImage.crop(invert(rgbImage).getbbox())
+        rgbImage = expand(rgbImage, border=10, fill=255)
 
         if arguments.debug:
             rgbImage.save(debugFilePath)
     # Get all previously scraped data # TODO: limit to filenames only
     alreadyScraped = ""
     for outputPath in [OUTPUT_PATH_PROFILE, OUTPUT_PATH_MOREINFO, OUTPUT_PATH_KILLS]:
-        with open(projectFolder + outputPath, "r+", newline='', encoding='utf-8') as outputFile:
+        with open(projectFolder + outputPath, "w+", newline='', encoding='utf-8') as outputFile:
             alreadyScraped = alreadyScraped + outputFile.read()
 
     # Mark as duplicates