In TextDetector.__init__() from detect_words.py,
The stored member variables of the TextDetector class have been updated to separate the two words on the communication module for discrete comparison with the detected text.
A new member variable has also been added to store the tesseract data generated. This is done for easy access of the data generated by tesseract for benchmarking and unit testing programs.
In TextDetector.detect_russian_word() from detect_words.py,
The text detection success parameters have been tweaked to only return BoundingBoxes that fulfil the requisite of having at least one whole word of the two on the communication module exist in the tesseract text data.
In TextDetector._get_rotated_min_area_rect() from detect_words.py,
The mask generated over the color image used to find the minimum area rectangle of the text has been updated to work with a a far wider range of brightness levels as it now uses a weighted ratio between the blue & green and blue & red channels.
In TextDetector._get_rotated_min_area_rect() from detect_words.py,
The function now consistently correctly rotates the detected minimum area rectangle.
Previously, the text detection would not work as intended on a Linux operating system due to OpenCV returning different a shifted value range for the cv2.minAreaRect() function. This is now corrected for Linux, but there is a commented out portion left for Windows developers and testers.
In BenchTextAccuracy.bench_accuracy() from bench_text.py,
Errors in the benchmark were now resolved.
The benchmark now outputs the text-detection-generated BoundingBoxes and full tesseract data to a file for each image file the benchmark processes to a "results" folder in the root directory of the file run.
Additionally, like before, the benchmark also can output marked image files of the detected text boxes to a "marked_images" folder in the root directory of the file run
In TextDetector.__init__() from detect_words.py, The stored member variables of the TextDetector class have been updated to separate the two words on the communication module for discrete comparison with the detected text. A new member variable has also been added to store the tesseract data generated. This is done for easy access of the data generated by tesseract for benchmarking and unit testing programs.
In TextDetector.detect_russian_word() from detect_words.py, The text detection success parameters have been tweaked to only return BoundingBoxes that fulfil the requisite of having at least one whole word of the two on the communication module exist in the tesseract text data.
In TextDetector._get_rotated_min_area_rect() from detect_words.py, The mask generated over the color image used to find the minimum area rectangle of the text has been updated to work with a a far wider range of brightness levels as it now uses a weighted ratio between the blue & green and blue & red channels.
In TextDetector._get_rotated_min_area_rect() from detect_words.py, The function now consistently correctly rotates the detected minimum area rectangle. Previously, the text detection would not work as intended on a Linux operating system due to OpenCV returning different a shifted value range for the cv2.minAreaRect() function. This is now corrected for Linux, but there is a commented out portion left for Windows developers and testers.
In BenchTextAccuracy.bench_accuracy() from bench_text.py, Errors in the benchmark were now resolved. The benchmark now outputs the text-detection-generated BoundingBoxes and full tesseract data to a file for each image file the benchmark processes to a "results" folder in the root directory of the file run. Additionally, like before, the benchmark also can output marked image files of the detected text boxes to a "marked_images" folder in the root directory of the file run