Hi, i'm actually trying to do the same thing that you've already tried on the same kind of images. Unfortunately with no results, i've even tried to train a neural network to recognize the numbers but even in that case with bad results. I know that have been passed a bit from your last post, but i really would like to know if you have found a solution.
Thanks in advance. Il giorno lunedì 11 aprile 2022 alle 01:37:29 UTC+2 javalover ha scritto: > > @zdenop: the code I posted in the @OP doesn't have do the rescaled image > and removed borders but I've already tried with no luck. > I excuse myself if I called you ignorant, but doesn't your sentence "It > does not matter how many times you post this (SO, issue tracker)..." feel a > little bit aggressive for no reason against me...? Even if I try to improve > quality, it doesn't mean it has to work at all. > :-) > I acknowledge that you may have contributed to the Tesseract project, but > could we please keep a non arrogant tone? I'm not familiar with Tesseract. > It's the first time I'm using it, maybe I'm doing something wrong. > Did you actually find the solution to my problem (does it work for you by > rescaling/removing borders)? If yes, I'd ask if you can post the code you > used, because it didn't work for me. > > Thanks. > Il giorno sabato 9 aprile 2022 alle 09:40:21 UTC+2 zdenop ha scritto: > >> Well >> https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md >> >> >> *Rescaling *-> Optimal resolution suggests that the size of input >> letters should be 30-33 points. YOU IGNORED it - your has 240 points!!! >> "for 4.x version use dark text on light background" -> YOU IGNORED IT. >> *Border *-> you did not remove them >> >> Conclusion: you are ignoring documentation. Good luck with asking for >> support while blaming others for ignorance. >> >> Zdenko >> >> >> pi 8. 4. 2022 o 0:56 javalover <[email protected]> napísal(a): >> >>> @zdenop: it doesn't matter how much arrogant/ignorant you can be, you >>> wouldn't have never understood that the reason I'm posting here is just >>> because of this >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#still-having-problems> >>> . >>> Thus, if you had actually read the page you linked me (Still having >>> problems? paragraph), you would have come to the conclusion that all >>> proposed solutions don't work: >>> >>> If you've tried the above and are still getting low accuracy results, ask >>> on the forum >>> <https://groups.google.com/forum/?fromgroups#!forum/tesseract-ocr> for >>> help, ideally posting an example image. >>> that page suggested to ask here for further problems. And as you've >>> already said, somebody has already suggested me to work on these image >>> processing procedures before feeding it to Tesseract: >>> >>> - Rescaling >>> >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#rescaling> >>> - Binarisation >>> >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#binarisation> >>> - Noise Removal >>> >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#noise-removal> >>> - Dilation / Erosion >>> >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#dilation-and-erosion> >>> - Rotation / Deskewing >>> >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#rotation--deskewing> >>> - Borders >>> >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#Borders> >>> - Transparency / Alpha channel >>> >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#transparency--alpha-channel> >>> - Tools / Libraries >>> >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#tools--libraries> >>> - Examples >>> >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#examples> >>> - Tables recognitions >>> >>> <https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md#tables-recognitions> >>> >>> So, I think you're smart enough to understand why I'm still here. Thanks. >>> >>> Il giorno giovedì 7 aprile 2022 alle 15:14:45 UTC+2 zdenop ha scritto: >>> >>>> It does not matter how many times you post this (SO, issue tracker) - >>>> answer will be same: read and follow instruction in >>>> https://github.com/tesseract-ocr/tessdoc/blob/main/ImproveQuality.md >>>> >>>> Zdenko >>>> >>>> >>>> št 7. 4. 2022 o 7:19 javalover <[email protected]> napísal(a): >>>> >>>>> I'm deskewing an image containing a number using Projection profile >>>>> based skew estimation algorithm >>>>> <http://www.cvc.uab.es/~bagdanov/pubs/ijdar98.pdf> and extracting it >>>>> through OCR. >>>>> >>>>> In order to calculate the correct skew angle, we compare the maximum >>>>> difference between peaks and using this skew angle, thus rotate the image >>>>> using the correct angle to correct the skew. >>>>> >>>>> Each image (which has a single number) has prefixed background and >>>>> foreground (number) color: >>>>> >>>>> - If the first pixel of the background is black, the foreground is >>>>> dark gray. >>>>> - If the first pixel of the background is white, the foreground is >>>>> dark white. >>>>> >>>>> Here are some sample images: >>>>> >>>>> [image: 1.jpg] [image: 2.jpg] [image: 3.jpg] [image: 4.jpg] >>>>> >>>>> which all of them get successfully deskewed to these: >>>>> >>>>> [image: 1.png] [image: 2.png] [image: 3.png] [image: 4.png] >>>>> >>>>> After it's been deskewed, I've tried with no luck to improve image >>>>> quality to let OCR (PyTesseract) recognize the numbers. >>>>> >>>>> >>>>> import cv2 >>>>> import numpy as np >>>>> import scipy.ndimage >>>>> import pytesseract >>>>> from PIL import Image, ImageEnhance, ImageFilter >>>>> >>>>> >>>>> from scipy.ndimage import interpolation as inter >>>>> >>>>> def correct_skew(image, delta=6, limit=150): >>>>> def determine_score(arr, angle): >>>>> data = inter.rotate(arr, angle, reshape=False, order=0) >>>>> histogram = np.sum(data, axis=1) >>>>> score = np.sum((histogram[1:] - histogram[:-1]) ** 2) >>>>> return histogram, score >>>>> >>>>> gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) >>>>> >>>>> blur = cv2.medianBlur(gray, 21) >>>>> >>>>> thresh = cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY | >>>>> cv2.THRESH_OTSU)[1] >>>>> >>>>> >>>>> scores = [] >>>>> angles = np.arange(-limit, limit + delta, delta) >>>>> for angle in angles: >>>>> histogram, score = determine_score(thresh, angle) >>>>> scores.append(score) >>>>> >>>>> best_angle = angles[scores.index(max(scores))] + 90 >>>>> >>>>> (h, w) = image.shape[:2] >>>>> center = (w // 2, h // 2) >>>>> M = cv2.getRotationMatrix2D(center, best_angle, 1.0) >>>>> rotated = cv2.warpAffine(image, M, (w, h), flags=cv2.INTER_CUBIC) >>>>> >>>>> >>>>> ## Image processing to improve OCR accuracy >>>>> """ >>>>> #rotated = cv2.medianBlur(rotated, 20) >>>>> >>>>> rotated = rotated.astype(np.float) / 255. >>>>> >>>>> # Calculate channel K: >>>>> rotated = 1 - np.max(rotated, axis=2) >>>>> >>>>> # Convert back to uint 8: >>>>> rotated = (255 * rotated).astype(np.uint8) >>>>> >>>>> binaryThresh = 190 >>>>> _, binaryImage = cv2.threshold(rotated, binaryThresh, 255, >>>>> cv2.THRESH_BINARY) >>>>> >>>>> binaryThresh = 190 >>>>> _, binaryImage = cv2.threshold(rotated, binaryThresh, 255, >>>>> cv2.THRESH_BINARY) >>>>> >>>>> # Use a little bit of morphology to clean the mask: >>>>> # Set kernel (structuring element) size: >>>>> kernelSize = 3 >>>>> # Set morph operation iterations: >>>>> opIterations = 2 >>>>> # Get the structuring element: >>>>> morphKernel = cv2.getStructuringElement(cv2.MORPH_RECT, >>>>> (kernelSize, kernelSize)) >>>>> # Perform closing: >>>>> rotated = cv2.morphologyEx(binaryImage, cv2.MORPH_CLOSE, >>>>> morphKernel, None, None, opIterations, cv2.BORDER_REFLECT101) >>>>> """ >>>>> >>>>> return best_angle, rotated >>>>> >>>>> if __name__ == '__main__': >>>>> image = cv2.imread('number.jpg') >>>>> if image[0][0][0] > 128: image = cv2.bitwise_not(image) >>>>> angle, rotated = correct_skew(image) >>>>> print(angle) >>>>> cv2.imshow('rotated', rotated) >>>>> cv2.imwrite('rotated.png', rotated) >>>>> pytesseract.pytesseract.tesseract_cmd = r'C:\\Program >>>>> Files\\Tesseract-OCR\\tesseract.exe' >>>>> >>>>> text = pytesseract.image_to_string(rotated, config="-c >>>>> tessedit_char_whitelist=0123456789") >>>>> print("number:", text) >>>>> cv2.waitKey() >>>>> >>>>> >>>>> This code (PyTesseract) recognizes the first >>>>> <https://i.stack.imgur.com/aPHQ3.png> and the second >>>>> <https://i.stack.imgur.com/ih5ei.png> number, while not the others. >>>>> Why? >>>>> >>>>> -- >>>>> 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 [email protected]. >>>>> To view this discussion on the web visit >>>>> https://groups.google.com/d/msgid/tesseract-ocr/b6d80f87-893b-40de-8067-a0cb1b0865d2n%40googlegroups.com >>>>> >>>>> <https://groups.google.com/d/msgid/tesseract-ocr/b6d80f87-893b-40de-8067-a0cb1b0865d2n%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 [email protected]. >>> >> To view this discussion on the web visit >>> https://groups.google.com/d/msgid/tesseract-ocr/9626b246-4983-439e-b2dc-24b7b77ec1d5n%40googlegroups.com >>> >>> <https://groups.google.com/d/msgid/tesseract-ocr/9626b246-4983-439e-b2dc-24b7b77ec1d5n%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 [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/25fbe5b8-23f1-4617-8b4e-81bb790ce40dn%40googlegroups.com.
