Tesseract is the OCR engine and it is not a text detection tool.
If you pass just blue button to tesseract, it has no problem to extract
text:
tesseract blue_button.png -
Sign in
Zdenko
št 22. 8. 2024 o 9:11 Abdul Kalam Shaik <shaikabdulkalam...@gmail.com>
napísal(a):
> Thanks Ger for your response. So, my use case is like when ever there is a
> colored background I'm unable to detect the text. Attached few use cases
> where I was facing difficulty in detecting the text.
>
> Regards,
>
> Shaik Abdul Kalam.
>
> On Tuesday, August 20, 2024 at 4:13:42 PM UTC+5:30 ger.h...@gmail.com
> wrote:
>
>> Generally, it is best to convert to greyscale with black text on white
>> background. Seems you tried that so questions remain.
>> Please include one or two sample images which exhibits your problem, so
>> folks around here have something to test against.
>>
>> Ciao,
>>
>> Ger
>>
>> On Mon, 19 Aug 2024, 18:45 Abdul Kalam Shaik, <shaikabdu...@gmail.com>
>> wrote:
>>
>>> Hello,
>>>
>>> I am encountering an issue with Tesseract OCR when trying to detect
>>> white text on a blue background. Despite various preprocessing techniques,
>>> the OCR is not accurately recognizing the text on this specific background.
>>>
>>> *Details:*
>>>
>>> Tesseract Version: tesseract v5.0.0-alpha.20210506
>>> Language Pack: English
>>> *Image Characteristics:*
>>> Background color: Blue
>>> Text color: White
>>> Image resolution: 1920X1080P
>>> Image format:PNG
>>> *Preprocessing Techniques Applied:*
>>> 1. Grayscale conversion
>>> 2. Contrast adjustment
>>> 3. Binary thresholding
>>> 4. Inversion of the image
>>> 5. Morphological operations
>>> 6. Increase Contrast
>>> 7. ROI
>>> 8. Convert the image to the HSV color space, Create a mask to isolate
>>> blue regions,Invert the mask to focus on the text and Using the mask to
>>> extract the white text
>>> * Script/Code Used:*
>>> import cv2
>>> import pytesseract
>>> import pyautogui
>>> import time
>>> import numpy as np
>>>
>>> # Specify the path to the Tesseract executable if not in PATH
>>> pytesseract.pytesseract.tesseract_cmd = r'C:\Program
>>> Files\Tesseract-OCR\tesseract.exe'
>>>
>>>
>>> def preprocess_image_gray(image):
>>> # Convert to grayscale
>>> gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
>>> cv2.imshow("Gray Scale Image", gray)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>> return gray
>>>
>>>
>>> def preprocess_image_increase_contrast(image):
>>> # Increase contrast
>>> contrast = cv2.convertScaleAbs(image, alpha=1.5, beta=0)
>>> cv2.imshow("Increase contrast", contrast)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>> return contrast
>>>
>>>
>>> def preprocess_image_gaussian_blur(image):
>>> # Apply Gaussian blur
>>> blurred = cv2.GaussianBlur(image, (5, 5), 0)
>>> cv2.imshow("GaussianBlur", blurred)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>> return blurred
>>>
>>>
>>> def preprocess_image_edge_detection(image):
>>> # Perform edge detection
>>> edged = cv2.Canny(image, 50, 150)
>>> cv2.imshow("edge detection", edged)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>> return edged
>>>
>>>
>>> def preprocess_image_inverted(image):
>>> # Invert the image
>>> inverted_image = cv2.bitwise_not(image)
>>> cv2.imshow("Inverted Image", inverted_image)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>>
>>> return inverted_image
>>>
>>>
>>> def preprocess_image_dialte_edges(image):
>>> # Dilate the edges
>>> dilated = cv2.dilate(image, None, iterations=2)
>>> cv2.imshow("dilate", dilated)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>>
>>> # Bitwise-AND mask and original image
>>> result = cv2.bitwise_and(image, image, mask=dilated)
>>> cv2.imshow("Bitwise-AND mask and original image", result)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>>
>>> # Invert the image
>>> inverted_image = cv2.bitwise_not(result)
>>> cv2.imshow("Inverted Image", inverted_image)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>>
>>> return inverted_image
>>>
>>>
>>> def perform_ocr(image_path, text_to_find=None, config="--psm 6 --oem 3",
>>> preprocess_func=preprocess_image_gray):
>>> global ocr_results
>>> try:
>>> image = cv2.imread(image_path)
>>> image_preprocessed = preprocess_func(image)
>>> image_rgb = cv2.cvtColor(image_preprocessed, cv2.COLOR_BGR2RGB)
>>> ocr_data = pytesseract.image_to_data(image_rgb,
>>> output_type=pytesseract.Output.DICT, config=config)
>>>
>>> if text_to_find is not None and not isinstance(text_to_find,
>>> list):
>>> text_to_find = [text_to_find]
>>>
>>> ocr_results = []
>>> for i in range(len(ocr_data['text'])):
>>> text = ocr_data['text'][i].strip()
>>> if not text:
>>> continue
>>>
>>> confidence = float(ocr_data['conf'][i]) / 100.0 # Convert
>>> confidence to decimal
>>> if confidence < 0.2: # Ignore results with confidence less
>>> than 0.5
>>> continue
>>>
>>> bbox = {
>>> "text": text,
>>> "left": ocr_data['left'][i],
>>> "right": ocr_data['left'][i] + ocr_data['width'][i],
>>> "top": ocr_data['top'][i],
>>> "bottom": ocr_data['top'][i] + ocr_data['height'][i],
>>> "confidence": float(ocr_data['conf'][i]) / 100.0 #
>>> Convert confidence to decimal
>>> }
>>> ocr_results.append(bbox)
>>> except Exception as e:
>>> print(f"An error occurred in the main function: {e}")
>>>
>>> return ocr_results
>>>
>>>
>>> def draw_boxes(image_path, ocr_results, output_image_path):
>>> image = cv2.imread(image_path)
>>>
>>> for result in ocr_results:
>>> x, y, w, h = result['left'], result['top'], result['right'] -
>>> result['left'], result['bottom'] - result['top']
>>> cv2.rectangle(image, (x, y), (x + w, y + h), (0, 255, 0), 2)
>>> distance_text = f"{result['text']} ({result['confidence']:.2f})"
>>> if 'distance' in result:
>>> distance_text += f" ({result['distance']:.2f})"
>>> cv2.putText(image, distance_text, (x, y - 10),
>>> cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
>>>
>>> cv2.imwrite(output_image_path, image)
>>>
>>>
>>> def increase_brightness(img, value=50):
>>> hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
>>> h, s, v = cv2.split(hsv)
>>>
>>> v = cv2.add(v, value)
>>> v[v > 255] = 255
>>> v[v < 0] = 0
>>>
>>> final_hsv = cv2.merge((h, s, v))
>>> brightened_img = cv2.cvtColor(final_hsv, cv2.COLOR_HSV2BGR)
>>> return brightened_img
>>>
>>>
>>> def isolate_white_text_on_blue(image):
>>> # Increase brightness
>>> brightened = increase_brightness(image, value=60)
>>> cv2.imshow("Brightened image", brightened)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>> # Convert the image to the HSV color space
>>> hsv = cv2.cvtColor(brightened, cv2.COLOR_BGR2HSV)
>>> cv2.imshow("HSV converted image", hsv)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>> # Define the range of blue colors in HSV
>>> lower_blue = np.array([100, 150, 0])
>>> upper_blue = np.array([140, 255, 255])
>>>
>>> # Create a mask to isolate blue regions
>>> blue_mask = cv2.inRange(hsv, lower_blue, upper_blue)
>>> cv2.imshow("Blue Mask image", blue_mask)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>> # Invert the mask to focus on the text
>>> blue_mask_inv = cv2.bitwise_not(blue_mask)
>>> cv2.imshow(" Inverted Mask image", blue_mask_inv)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>> # Use the mask to extract the white text
>>> white_text_on_blue = cv2.bitwise_and(brightened, brightened,
>>> mask=blue_mask_inv)
>>> cv2.imshow("White Text image", blue_mask_inv)
>>> cv2.waitKey(0)
>>> cv2.destroyAllWindows()
>>>
>>> return white_text_on_blue
>>>
>>>
>>> def move_and_click(ocr_results):
>>> for result in ocr_results:
>>> x_center = (result['left'] + result['right']) // 2
>>> y_center = (result['top'] + result['bottom']) // 2
>>> pyautogui.moveTo(x_center, y_center)
>>> pyautogui.click()
>>> time.sleep(1) # Sleep for a second between clicks for safety
>>>
>>>
>>> def main():
>>> image_path = 'path-to-image.png'
>>> output_json_path = 'path-to-image.json'
>>> output_image_path = 'path-to-outputimage.png'
>>>
>>> text_to_find = []
>>>
>>> ocr_results_gray = perform_ocr(image_path, text_to_find,
>>> config="--psm 11 --oem 3",
>>> preprocess_func=preprocess_image_gray)
>>> print(f"OCR gray results are: {ocr_results_gray}")
>>> ocr_results_contrast = perform_ocr(image_path, text_to_find,
>>> config="--psm 11 --oem 3",
>>>
>>> preprocess_func=preprocess_image_increase_contrast)
>>> print(f"OCR contrast results are: {ocr_results_contrast}")
>>> ocr_results_gaussian = perform_ocr(image_path, text_to_find,
>>> config="--psm 11 --oem 3",
>>>
>>> preprocess_func=preprocess_image_gaussian_blur)
>>> print(f"OCR gaussian results are: {ocr_results_gaussian}")
>>> ocr_results_edge = perform_ocr(image_path, text_to_find,
>>> config="--psm 11 --oem 3",
>>>
>>> preprocess_func=preprocess_image_edge_detection)
>>> print(f"OCR edge results are: {ocr_results_edge}")
>>> ocr_results_dialte = perform_ocr(image_path, text_to_find,
>>> config="--psm 11 --oem 3",
>>>
>>> preprocess_func=preprocess_image_dialte_edges)
>>> print(f"OCR Dialte results are: {ocr_results_dialte}")
>>> ocr_results_invert = perform_ocr(image_path, text_to_find,
>>> config="--psm 11 --oem 3",
>>>
>>> preprocess_func=preprocess_image_inverted)
>>> print(f"OCR Invert results are: {ocr_results_invert}")
>>> ocr_results_isolate = perform_ocr(image_path, text_to_find,
>>> config="--psm 11 --oem 3",
>>>
>>> preprocess_func=isolate_white_text_on_blue)
>>> print(f"OCR Isolate results are: {ocr_results_isolate}")
>>>
>>> ocr_results = []
>>> if isinstance(ocr_results_gray, list) and
>>> isinstance(ocr_results_isolate, list) and isinstance(ocr_results_invert,
>>>
>>> list) and isinstance(
>>> ocr_results_contrast, list) and
>>> isinstance(ocr_results_gaussian, list) and isinstance(ocr_results_dialte,
>>>
>>> list) and isinstance(
>>> ocr_results_edge, list):
>>> ocr_results = ocr_results_isolate + ocr_results_gray +
>>> ocr_results_contrast + ocr_results_gaussian + ocr_results_dialte +
>>> ocr_results_edge + ocr_results_invert
>>> else:
>>> print("OCR results are not in the expected list format.")
>>>
>>>
>>> for i, result in enumerate(ocr_results, start=1):
>>> bounding_box_info = (
>>> f"Bounding box: Text = {result['text']}, Left =
>>> {result['left']}, Top = {result['top']}, "
>>> f"Right = {result['right']}, Bottom = {result['bottom']},
>>> Confidence = {result['confidence']:.2f}"
>>> )
>>> print(bounding_box_info)
>>> print()
>>>
>>> draw_boxes(image_path, ocr_results, output_image_path)
>>>
>>> print(ocr_results)
>>>
>>>
>>> if __name__ == "__main__":
>>> main()
>>>
>>> *Issue:*
>>> Despite trying the above preprocessing techniques, the OCR output is
>>> still missing or incorrectly recognizing the text on the blue background. I
>>> have also tried adjusting the thresholding and brightness levels, but
>>> without success.
>>>
>>> *Question:*
>>> Could anyone provide suggestions on additional preprocessing techniques
>>> or modifications to the Tesseract OCR settings that might help improve the
>>> detection accuracy for white text on a blue background or any other colored
>>> background?
>>>
>>> Thank you in advance for your assistance!
>>>
>>> --
>>> You received this message because you are subscribed to the Google
>>> Groups "tesseract-ocr" group.
>>> To unsubscribe from this group and stop receiving emails from it, send
>>> an email to tesseract-oc...@googlegroups.com.
>>> To view this discussion on the web visit
>>> https://groups.google.com/d/msgid/tesseract-ocr/0ce2e54e-1f1a-4fca-8c74-286c9641509en%40googlegroups.com
>>> <https://groups.google.com/d/msgid/tesseract-ocr/0ce2e54e-1f1a-4fca-8c74-286c9641509en%40googlegroups.com?utm_medium=email&utm_source=footer>
>>> .
>>>
>> --
> You received this message because you are subscribed to the Google Groups
> "tesseract-ocr" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to tesseract-ocr+unsubscr...@googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/tesseract-ocr/6c7383e8-81a1-4259-9df5-1185ad84b077n%40googlegroups.com
> <https://groups.google.com/d/msgid/tesseract-ocr/6c7383e8-81a1-4259-9df5-1185ad84b077n%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>
--
You received this message because you are subscribed to the Google Groups
"tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to tesseract-ocr+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8zJi9TyZh6co%2BkD9ik0ARvGH0f7XGN__X1oxxVU3THnxA%40mail.gmail.com.
