如何使用 PyTesseract 去除图像噪声以改善结果？

Question

我正在尝试从视频的左上角获取文本“P1”和“P2”。

P1

P2

我拍摄一帧并将它们裁剪为以下图像，然后应用此处找到的图像处理：

P1 Crop

P2 Crop

use pytesseract to recognize text from image

虽然它适用于我使用图像编辑器手动编辑的裁剪静止图像，但在使用 cv2 从视频中获取帧时它不起作用。

我不确定这是为什么，但我怀疑它与下图中的黑白背景有关，但我不知道如何在不删除文本的情况下摆脱它。

P1 post image manipulation

这是我的代码

import cv2
import pytesseract
import re
from difflib import SequenceMatcher
def determineWinner(video):
    winnerRect = [(70,95),(146,152)]
    cap = cv2.VideoCapture(video)
    if(cap.isOpened() == False):
        print("No dice")
        return
    fps = cap.get(cv2.CAP_PROP_FPS)
    frames = cap.get(cv2.CAP_PROP_FRAME_COUNT)
    print(fps)
    print(frames)
    desiredSeek = frames - int(fps * 9)
    print(desiredSeek)
    seconds = desiredSeek/fps
    print(seconds)
    minutes = seconds/60
    print(minutes)
    partial = minutes - int(minutes)
    print(partial)
    seconds = partial * 60
    print(seconds)
    print(str(int(minutes)) +":"+ str(seconds))

    cap.set(cv2.CAP_PROP_POS_FRAMES,(desiredSeek))

    ret,img = cap.read()
    winTxt = []
    p1Count = 0
    p2Count = 0

    cv2.namedWindow("",cv2.WINDOW_NORMAL)

    ret,img = cap.read()
    while ret:
        key = cv2.waitKey(1)
        if key == ord('q'):
            break
        if key == ord('e'):
            ret,img = cap.read()
            if ret:
                winROI = img[winnerRect[0][1]:winnerRect[1][1],winnerRect[0][0]:winnerRect[1][0]]
                gray = cv2.cvtColor(winROI, cv2.COLOR_BGR2GRAY)
                blur = cv2.GaussianBlur(gray, (3,3), 0)
                thresh = cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]

                # Morph open to remove noise and invert image
                kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3,3))
                opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=1)
                invert = 255-opening
                invert=cv2.resize(invert,None,fx=2,fy=2)
                wConfig='-l eng --oem 1 --psm 10 -c tessedit_char_whitelist=P12'
                winTxt = pytesseract.image_to_string(invert,config=wConfig)
                cv2.rectangle(img,winnerRect[0],winnerRect[1],(255,0,0),2)
                cv2.imshow("winroi",invert)
                cv2.imshow("",img)
                cv2.resizeWindow("",800,600)
                print(winTxt)
                desiredSeek+=1
                seconds = desiredSeek/fps
                minutes = seconds/60
                partial = minutes - int(minutes)
                seconds = partial * 60
                print(str(int(minutes)) +":"+ str(seconds))
            else:
                break

    cap.release()
    cv2.destroyAllWindows()

Answer 1

此代码用作测试脚本。我只提取了包含P1 的图像的参数。要在新图像上应用过滤器，只需擦除预定义的阈值，如下所示：

发件人：

low_blue, low_green, low_red, upper_blue, upper_green, upper_red = (115, 0, 0, 255, 178, 255)

收件人：

low_blue, low_green, low_red, upper_blue, upper_green, upper_red = (0, 0, 0, 255, 255, 255)

并开始修改参数，如下所述。确定好参数后，按esc退出程序，取控制台显示的参数，粘贴到thresholds元组中。

---

如何使用：

• 非常重要。要使其正常工作，您必须单击鼠标左键选择来自 cv2.imshow() 的窗口，在本例中为 Original image 或 Binary image

• q 增加，w 降低蓝色下限

• a 增加，s 降低绿色阈值下限
• ...对于上下颜色 (BGR) 阈值，依此类推

---

import numpy as np
import cv2

low_blue, low_green, low_red, upper_blue, upper_green, upper_red = (115, 0, 0, 255, 178, 255)

# Get picture
path = "C:\\Users\\asd\\asd\\P1.png"
frame = cv2.imread(path)

while 1:

    lower_color = np.array((low_blue, low_green, low_red))
    upper_color = np.array((upper_blue, upper_green, upper_red))

    # extract binary image with active blue regions
    binary_image = cv2.inRange(frame, lower_color, upper_color)

    cv2.imshow('Original image', binary_image)

    #erode for the little white contour to dissapear
    binary_image = cv2.erode(binary_image, cv2.getStructuringElement(cv2.MORPH_RECT,(3,3)))
    binary_image = cv2.dilate(binary_image, cv2.getStructuringElement(cv2.MORPH_RECT,(3,3)))

    cv2.imshow('Binary image  ', binary_image)

    k = cv2.waitKey(5) & 0xFF
    if k == 27:
        break
    if k == ord('q'):
        low_blue += 1
    if k == ord('w'):
        low_blue -= 1
    if k == ord('a'):
        low_green += 1
    if k == ord('s'):
        low_green -= 1
    if k == ord('z'):
        low_red += 1
    if k == ord('x'):
        low_red -= 1
    if k == ord('e'):
        upper_blue += 1
    if k == ord('r'):
        upper_blue -= 1
    if k == ord('d'):
        upper_green += 1
    if k == ord('f'):
        upper_green -= 1
    if k == ord('c'):
        upper_red += 1
    if k == ord('v'):
        upper_red -= 1

    print("low_blue=", low_blue, "low_green=", low_green, "low_red=",low_red, "upper_blue", upper_blue, "upper_green=",
          upper_green, "upper_red=",upper_red)


cv2.destroyAllWindows()

---

结果

发件人：

收件人：