Implement text detection based on tesseract's boxes extension

This allows us to use fewer backends and dependencies for relatively
well performing OCR backend configurations, in this case just
pytesseract for both text detection and text recognition (OCR).

guibot/finder.py

@@ -1799,7 +1799,7 @@ def __init__(self, configure=True, synchronize=True):
         self.categories["threshold2"] = "threshold_filters2"
         self.categories["threshold3"] = "threshold_filters3"
         self.algorithms["text_matchers"] = ("mixed",)
-        self.algorithms["text_detectors"] = ("east", "erstat", "contours", "components")
+        self.algorithms["text_detectors"] = ("pytesseract", "east", "erstat", "contours", "components")
         self.algorithms["text_recognizers"] = ("pytesseract", "tesserocr", "tesseract", "hmm", "beamSearch")
         self.algorithms["threshold_filters2"] = tuple(self.algorithms["threshold_filters"])
         self.algorithms["threshold_filters3"] = tuple(self.algorithms["threshold_filters"])
@@ -1858,7 +1858,17 @@ def __configure_backend(self, backend=None, category="text", reset=False):
         if category == "text":
             self.params[category]["datapath"] = CVParameter("../misc")
         elif category == "tdetect":
-            if backend == "east":
+            if backend == "pytesseract":
+                # eng, deu, etc. (ISO 639-3)
+                self.params[category]["language"] = CVParameter("eng")
+                self.params[category]["char_whitelist"] = CVParameter(" 0123456789abcdefghijklmnopqrst"
+                                                                      "uvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ")
+                # 0 original tesseract only, 1 neural nets LSTM only, 2 both, 3 anything available
+                self.params[category]["oem"] = CVParameter(3, 0, 3, enumerated=True)
+                # 13 different page segmentation modes - see Tesseract API
+                self.params[category]["psmode"] = CVParameter(3, 0, 13, enumerated=True)
+                self.params[category]["extra_configs"] = CVParameter("")
+            elif backend == "east":
                 # network input dimensions - must be divisible by 32, however currently only
                 # 320x320 doesn't error out from the OpenCV implementation
                 self.params[category]["input_res_x"] = CVParameter(320, 32, None, 32.0)
@@ -1994,6 +2004,20 @@ def __synchronize_backend(self, backend=None, category="text", reset=False):
             # nothing to sync
             return
 
+        elif category == "tdetect" and backend == "pytesseract":
+            tessdata_path = os.path.join(datapath, "tessdata")
+            if not os.path.exists(tessdata_path):
+                tessdata_path = os.environ.get("TESSDATA_PREFIX", ".")
+
+            import pytesseract
+            self.tbox = pytesseract
+            self.tbox_config = r"--tessdata-dir %s --oem %s --psm %s "
+            self.tbox_config %= (tessdata_path,
+                                 self.params["tdetect"]["oem"].value,
+                                 self.params["tdetect"]["psmode"].value)
+            self.tbox_config += r"-c tessedit_char_whitelist='%s' %s batch.nochop wordstrbox"
+            self.tbox_config %=  (self.params["tdetect"]["char_whitelist"].value,
+                                  self.params["tdetect"]["extra_configs"].value)
         elif category == "tdetect" and backend == "east":
             self.east_net = cv2.dnn.readNet(os.path.join(datapath, 'frozen_east_text_detection.pb'))
         elif category == "tdetect" and backend == "erstat":
@@ -2132,7 +2156,9 @@ def find(self, needle, haystack):
         # detect characters and group them into detected text
         backend = self.params["tdetect"]["backend"]
         log.debug("Detecting text with %s", backend)
-        if backend == "east":
+        if backend == "pytesseract":
+            text_regions = self._detect_text_boxes(haystack)
+        elif backend == "east":
             text_regions = self._detect_text_east(haystack)
         elif backend == "erstat":
             text_regions = self._detect_text_erstat(haystack)
@@ -2243,6 +2269,36 @@ def binarize_step(threshold, text_img):
         self.imglog.log(30)
         return matches
 
+    def _detect_text_boxes(self, haystack):
+        import cv2
+        import numpy
+        img_haystack = numpy.array(haystack.pil_image)
+        img_haystack = numpy.array(haystack.pil_image)
+        self.imglog.hotmaps.append(img_haystack)
+
+        output = self.tbox.run_and_get_output(img_haystack, 'box',
+                                              self.params["tdetect"]["language"].value,
+                                              config=self.tbox_config)
+
+        text_regions = []
+        for line in output.splitlines():
+            tokens = line.rstrip().split(" ", maxsplit=6)
+            if tokens[0] != "WordStr":
+                continue
+            left = int(tokens[1])
+            bottom = haystack.height - int(tokens[2])
+            right = int(tokens[3])
+            top = haystack.height - int(tokens[4])
+            text = tokens[6][1:]
+
+            x, y, w, h = left, top, right - left, bottom - top
+            logging.debug("Found text '%s' with tesseract-provided box %s", text, (x, y, w, h))
+            cv2.rectangle(img_haystack, (x, y), (x+w, y+h), (0, 0, 0), 2)
+            cv2.rectangle(img_haystack, (x, y), (x+w, y+h), (0, 255, 0), 1)
+            text_regions.append([x, y, w, h])
+
+        return text_regions
+
     def _detect_text_east(self, haystack):
         #:.. note:: source implementation by Adrian Rosebrock from his post:
         #:   https://www.pyimagesearch.com/2018/08/20/opencv-text-detection-east-text-detector/
@@ -2284,7 +2340,7 @@ def _detect_text_east(self, haystack):
                 w = min(row_data[1][col] + row_data[3][col], inp_width) * width_ratio
                 # output layer dimensions are 4x smaller than the input layer dimentions
                 (dx, dy) = (col + 1) * 4.0, (row + 1) * 4.0
-                # calculate the rotation angle from the prediction ouput
+                # calculate the rotation angle from the prediction output
                 sin, cos = numpy.sin(row_data[4][col]), numpy.cos(row_data[4][col])
                 # compute the starting (from ending) coordinates for the text bounding box
                 x2 = min(dx + cos * row_data[1][col] + sin * row_data[2][col], inp_width) * width_ratio