绘制纯色图片

import cv2
import numpy as np
# 512*512*3 长*宽*通道数
white_img = np.ones((512,512,3), np.uint8)
# 每个元素设置为 255 => (255,255,255) 纯白
white_img = 255*white_img 
# 显示图片
cv2.imshow('white_img', white_img)
# 等待回车按键
if cv2.waitKey(0) == 27:
    cv2.destroyAllWindows()

常用绘制函数

import cv2
import numpy as np

img = np.ones((512, 512, 3), np.uint8)
img = 255 * img
# 直线绘制函数 cv2.line(image, starting, ending, color, thickness, lineType)
img = cv2.line(img, (100, 100), (400, 400), (255, 0, 0), 5)
# 长方形绘制函数 cv2.rectangle(image, top-left, bottom-right, color, thickness, lineType)
img = cv2.rectangle(img, (200, 20), (400, 120), (0, 255, 0), 3)
# 圆形绘制函数 cv2.circle(image, center, radius, color, thickness, lineType)
img = cv2.circle(img, (100, 400), 50, (0, 0, 255), 2)
img = cv2.circle(img, (250, 400), 50, (0, 0, 255), 0)
# 椭圆绘制函数 cv2.circle(image, center, (major-axis-length, minor-axis-length), angle, startAngle, endAngle, color, thickness, lineType)
img = cv2.ellipse(img, (256, 256), (100, 50), 0, 0, 180, (0, 255, 255), -1)
# 多边形绘制函数 cv2.polylines(image, [point-set], flag, color, thickness, lineType)
pts = np.array([[10, 5], [20, 30], [70, 20], [50, 10]], np.int32)
img = cv2.polylines(img, [pts], True, (0, 0, 0), 2)

cv2.imshow('img', img)
if cv2.waitKey(0) == 27:
    cv2.destroyAllWindows()

图片简单处理

import cv2
import imutils
import numpy as np

rgb_img = cv2.imread('C:/Users/youhl002/Downloads/678702.jpg')
# 缩放图片
rgb_img = imutils.resize(rgb_img, 300)
# 原图
cv2.imshow('rgb_img', rgb_img)
# 灰度
gray_img = cv2.cvtColor(rgb_img, cv2.COLOR_BGR2GRAY)
cv2.imshow('gray_img', gray_img)
# 灰度取反
reverse_img = 255 - gray_img
cv2.imshow('reverse_img', reverse_img)
# 线性处理
linear_img = np.zeros((gray_img.shape[0], gray_img.shape[1]), dtype=np.uint8)
for i in range(gray_img.shape[0]):
    for j in range(gray_img.shape[1]):
        linear_img[i, j] = gray_img[i, j] * 1.2
cv2.imshow('linear_img', linear_img)
if cv2.waitKey(0) == 27:
    cv2.destroyAllWindows()

银行卡号识别

import cv2
import numpy as np


def get_contours_rets(contours):
    rects = []
    for i, cnt in enumerate(contours):
        x1, y1, w1, h1 = cv2.boundingRect(cnt)
        rects.append([x1, y1, w1, h1])
    rects.sort(key=lambda x: x[0])
    return rects


# 归一化数字
def number_normalize(img, rect, pad=0, size=(57, 88)):
    x, y, w, h = rect
    original = img[y - pad:y + h + pad, x - pad:x + w + pad]
    return cv2.resize(original, size)


# 计算分数
def get_scores(targets, source):
    scores = []
    for t in targets:
        result = cv2.matchTemplate(source, t, cv2.TM_CCOEFF)
        (min_val, max_val, min_loc, max_loc) = cv2.minMaxLoc(result)
        scores.append(max_val)
    return scores


class Ocr:
    def __init__(self, ot_img: np.ndarray, src_img: np.ndarray):
        self.ot_img: np.ndarray = ot_img
        self.src_img: np.ndarray = src_img
        self.src_gray: np.ndarray = None
        self.labels = []
        self.rects = []
        print(f"图片尺寸 {src_img.shape}")

    def parse(self):
        self.parse_template()
        self.parse_src()
        card = []
        for rect in self.rects:
            x, y, w, h = rect
            # 标记识别的数字轮廓
            cv2.rectangle(self.src_img, (x - 10, y - 5), (x + w + 3, y + h + 2), (0, 0, 255), 1)
            # 搜索单个数字轮廓
            box = self.src_gray[y - 5:y + h + 5, x - 5:x + w + 5]
            box_thresh = cv2.threshold(box, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
            contours = cv2.findContours(box_thresh.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[0]
            cv2.drawContours(box_thresh, contours, -1, (127, 255,), 3)
            num_boxes = get_contours_rets(contours)
            numbers = []
            for num_box in num_boxes:
                roi = number_normalize(box_thresh, num_box, 1)
                scores = get_scores(self.labels, roi)
                # 返回最大值的索引, 因为模板是0-9所以索引就是目标值, 否则还需要进一步获取原始数据值
                index = np.argmax(scores)
                numbers.append(index)
                card.append(index)
            # 在图中标记数字
            cv2.putText(src_img, "".join([str(v) for v in numbers]), (x - 10, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.8,
                        (0, 0, 255), 2)
        print(f"识别的银行卡号 => {card}")

    def parse_template(self):
        # 灰度图片
        ot_gray = cv2.cvtColor(self.ot_img, cv2.COLOR_BGR2GRAY)
        # 模板灰度图片二值化
        ot_bit = cv2.threshold(ot_gray, 10, 255, cv2.THRESH_BINARY_INV)[1]
        # 根据二值图找轮廓
        contours = cv2.findContours(ot_bit, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[0]
        labels = get_contours_rets(contours)
        print(f"模板轮廓 {labels}")
        # 绘制轮廓
        # ot_bit_rect = ot_bit.copy()
        # cv2.drawContours(ot_bit_rect, contours, -1, (127, 255,), 3)
        # cv2.imshow('ot_img', ot_img)
        # cv2.imshow('template', np.vstack([ot_gray, ot_bit, ot_bit_rect]))
        self.labels = [number_normalize(ot_bit, r) for r in labels]

    def parse_src(self):
        src_card = self.src_img.copy()
        # 将 [0:0.5h] 和 [0.66h,1h] 范围的像素设置为0
        src_card[0:src_card.shape[0] // 2] = 0
        src_card[2 * src_card.shape[0] // 3:] = 0
        print(f"截取高度范围 ({src_card.shape[0] // 2}:{2 * src_card.shape[0] // 3})")
        # 转换为灰度图片
        self.src_gray = cv2.cvtColor(src_card, cv2.COLOR_BGR2GRAY)
        # sobel化 (索贝尔算子主要用作边缘检测), ksize=-1相当于用3*3的
        src_sobel = cv2.Sobel(self.src_gray, ddepth=cv2.CV_64F, dx=1, dy=0, ksize=-1)
        # 对每个元素求绝对值
        src_sobel = np.absolute(src_sobel)
        (minVal, maxVal) = (np.min(src_sobel), np.max(src_sobel))
        # 全局按照 255的区间重新格式化
        src_sobel = (255 * ((src_sobel - minVal) / (maxVal - minVal)))
        # 重新转换为整形
        src_sobel = np.uint8(src_sobel)
        # 创建内核矩阵
        # 利用 morphologyEx 形态学变化函数进行滤波操作 (cv2.MORPH_CLOSE 闭运算,先膨胀后腐蚀)
        # 利用 threshold 固定阈值操作, 去除噪声, 过滤掉像素值很小或很大的图像点。(THRESH_BINARY|THRESH_OTSU：使用最大类间差分法（THRESH_OTSU）获取阈值，然后再用该阈值进行二分。)
        # 再次利用 morphologyEx 形态学变化函数进行滤波操作
        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (20, 3))
        src_close = cv2.morphologyEx(src_sobel, cv2.MORPH_CLOSE, kernel)
        src_thresh = cv2.threshold(src_close, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
        src_close = cv2.morphologyEx(src_thresh, cv2.MORPH_CLOSE, kernel)
        # 在处理好的图片上寻找轮廓 并 绘制
        contours = cv2.findContours(src_close.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[0]
        src_card_1 = src_card.copy()
        cv2.drawContours(src_card_1, contours, -1, (0, 0, 255), 3)
        # 记录抓取的轮库
        rects = []
        for cnt in contours:
            if cv2.contourArea(cnt) > 1000:  # 防止抓取的区域过小
                L = cv2.arcLength(cnt, True)
                approx = cv2.approxPolyDP(cnt, 0.02 * L, True)
                x, y, w, h = cv2.boundingRect(approx)
                rects.append([x, y, w, h])
        rects.sort(key=lambda v: v[0])
        print(f"图片轮廓 {rects}")
        # cv2.imshow('src_card', np.vstack([src_img, src_card, src_card_1]))
        # cv2.imshow('src_gray', np.vstack([self.src_gray, src_sobel]))
        # cv2.imshow('src_thresh', np.vstack([src_thresh, src_close]))
        self.rects = rects

if __name__ == '__main__':
    ot_img = cv2.imread('ocr_a_reference.png')
    src_img = cv2.imread('bank.png')
    ocr = Ocr(ot_img, src_img)
    ocr.parse()
    cv2.imshow('ocr', np.vstack([ot_img]))
    cv2.imshow('out', np.vstack([src_img]))
    cv2.waitKey(0)
    cv2.destroyAllWindows()