OpenCV+Python 文字識別（重點圖像透視變換）

# Author：Winter Liu is coming!
import cv2 as cv
import numpy as np
import pytesseract


# 預處理，高斯濾波（用處不大），4次開操作
# 過濾輪廓唯一
def contour_demo(img):
    gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
    gray = cv.GaussianBlur(gray, (5, 5), 1)
    ref, thresh = cv.threshold(gray, 127, 255, cv.THRESH_BINARY)
    kernel = np.ones((9, 9), np.uint8)
    thresh = cv.morphologyEx(thresh, cv.MORPH_OPEN, kernel, iterations=4)
    contours, hierachy = cv.findContours(thresh, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
    print(len(contours))
    return contours


def capture(img):
    contours = contour_demo(img)
    # 輪廓唯一，以后可以擴展
    contour = contours[0]
    # 求周長，可在后面的轉換中使用周長和比例
    print(cv.arcLength(contour,True))
    img_copy = img.copy()
    # 使用approxPolyDP，將輪廓轉換為直線，22為精度（越高越低），TRUE為閉合
    approx = cv.approxPolyDP(contour, 22, True)
    # print(approx.shape)
    # print(approx)
    # cv.drawContours(img_copy, [approx], -1, (255, 0, 0), 15)
    n = []
    # 生產四個角的坐標點
    for x, y in zip(approx[:, 0, 0], approx[:, 0, 1]):
        n.append((x, y))
    p1 = np.array(n, dtype=np.float32)
    # 對應點
    p2 = np.array([(0, 0), (0, 1500), (1000, 1500), (1000, 0)], dtype=np.float32)
    M = cv.getPerspectiveTransform(p1, p2) # 變換矩陣
    # 使用透視變換
    result = cv.warpPerspective(img_copy, M, (0, 0))
    # 重新截取
    result = result[:1501, :1001]
    cv.imwrite(r"C:\PycharmProjects\OpenCV\pic\ocr.png", result)
    return result


# 圖像識別代碼，需要預先下載安裝開源工具包 pytesseract，配置環境變量
# pip install pytesseract
# 修改“C:\Python\Python37\Lib\site-packages\pytesseract\pytesseract.py”中“cmd”為絕對路徑
def ocr_img(img):
    gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
    # 圖像清晰度越高結果越精確，時間更長
    text = pytesseract.image_to_string(gray)
    print(text)


src = cv.imread(r"C:\PycharmProjects\OpenCV\pic\page.jpg")
res = capture(src)
ocr_img(res)
cv.waitKey(0)
cv.destroyAllWindows()