python+opencv生成较真实的车牌号码图片

本文参考github代码：https://github.com/loveandhope/license-plate-generator

效果：

一、代码目录结构：

background目录下存放各种背景图片

font目录下存放车牌中文、字符的ttf字体

images目录下存放蓝色底牌、新能源绿色底牌、污渍（噪声）的图片

完整代码可参考：https://download.csdn.net/download/benben044/87546578?spm=1001.2014.3001.5503

二、生成流程

本代码可以根据车牌list生成对应的车牌图片list。

（1）生成白底黑字的车牌号码图片

首先，生成一个白底的空白车牌号图片

img = np.array(Image.new("RGB", (880, 280), (255, 255, 255)))

然后，逐字（字符）生成图片

ImageDraw.Draw(img).text((0, self.height_offset), char, self.fg_color, font=self.font_en)

最后，将上一步生成的图片逐个复制到上上一步步生成的图片中

img[:, char_width_start:char_width_end] = self.generate_char_image(plate_num[i])

因为生成的图片和img的第一维（height）大小相同，所以在img中直接使用符号":"。

（2）生成车牌底牌

直接读取底牌的图片即可。

plate_image = cv2.imread(LicensePlateImageGenerator.single_blue_plate_bg)

（3）生成最后的车牌图片（以蓝牌为例）

首先，将文字图片转为黑底白字

img = cv2.bitwise_not(char_img)

此时，背景部分值为0，字部分值为255。

然后，将黑底背景变为蓝色底牌背景

img = cv2.bitwise_or(img, template_image)

此时，黑色背景值0 与蓝色背景值x进行二进制的or操作，只保留了蓝色背景值，实现了背景的替换。

（4）数据增强（增加噪声）

高斯模糊：

通过cv2.blur() 方法实现

高斯噪声：

    def add_single_channel_noise(self, single):""" 添加高斯噪声:param single: 单一通道的图像数据"""diff = 255 - single.max()noise = np.random.normal(0, 1 + self.rand_reduce(6), single.shape)noise = (noise - noise.min()) / (noise.max() - noise.min())noise = diff * noisenoise = noise.astype(np.uint8)dst = single + noisereturn dst

添加污渍：

通过cv2.bitwise_not() 和cv2.bitwise_and()操作完成

添加饱和度光照的噪声：

通过调整HSV颜色空间实现，

Hue：色调
Saturation：饱和度
Value：明亮度

hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
# 色调，饱和度，亮度
hsv[:, :, 0] = hsv[:, :, 0] * (self.hue_keep + np.random.random() * (1 - self.hue_keep))
hsv[:, :, 1] = hsv[:, :, 1] * (self.saturation_keep + np.random.random() * (1 - self.saturation_keep))
hsv[:, :, 2] = hsv[:, :, 2] * (self.value_keep + np.random.random() * (1 - self.value_keep))
img = cv2.cvtColor(hsv, cv2.COLOR_HSV2BGR)

添加透视变换：

透视变换api整体和放射变化类似，先通过点与点的映射关系获取变换矩阵，然后再将图形进行转换。

以左向投影为例：

其计算过程如下：

寻找源图像中4个点和目标图像中的4个点，在左向倾斜时右边两个顶点的height值会做修改。

shape = img.shape
size_src = (shape[1], shape[0])  # width, height
# 源图像四个顶点坐标
pts1 = np.float32([[0, 0], [0, size_src[1]], [size_src[0], 0], [size_src[0], size_src[1]]])
# 计算图片进行投影倾斜后的位置
interval = abs(int(math.sin((float(angle) / 180) * math.pi) * shape[0]))
# 目标图像上四个顶点的坐标
if is_left:pts2 = np.float32([[0, 0], [0, size_src[1]],[size_src[0], interval], [size_src[0], size_src[1] - interval]])
else:pts2 = np.float32([[0, interval], [0, size_src[1] - interval],[size_src[0], 0], [size_src[0], size_src[1]]])
# 获取 3x3的投影映射/透视变换 矩阵
matrix = cv2.getPerspectiveTransform(pts1, pts2)
dst = cv2.warpPerspective(img, matrix, size_src)

三、python代码：

import os
from PIL import ImageFont
from PIL import Image
from PIL import ImageDraw
import numpy as np
import cv2
import random
import mathclass CharsImageGenerator(object):"""生成字符图像，背景为白色，字体为黑色"""# 数字和英文字母列表numerals = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']alphabet = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J','K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T','U', 'V', 'W', 'X', 'Y', 'Z']def __init__(self, plate_type):self.plate_type = plate_type# 字符图片参数self.font_ch = ImageFont.truetype("./font/platech.ttf", 180, 0)  # 中文字体格式self.font_en = ImageFont.truetype('./font/platechar.ttf', 240, 0)  # 英文字体格式self.bg_color = (255, 255, 255)  # 车牌背景颜色self.fg_color = (0, 0, 0)  # 车牌号的字体颜色self.plate_height = 280  # 车牌高度self.left_offset = 32  # 车牌号左边第一个字符的偏移量self.height_offset = 10  # 高度方向的偏移量self.char_height = 180  # 字符高度self.chinese_original_width = 180  # 中文字符原始宽度self.english_original_width = 90  # 非中文字符原始宽度if plate_type in ['single_blue', 'single_yellow']:self.char_num = 7self.char_width = 90  # 字符校正后的宽度self.plate_width = 880  # 车牌的宽度self.char_interval = 24  # 字符间的间隔self.point_size = 20  # 第2个字符与第三个字符间有一个点，该点的尺寸elif plate_type == 'small_new_energy':self.char_num = 8self.first_char_width = 90  # 第一个字符校正后的宽度self.char_width = 86  # 其余字符校正后宽度self.plate_width = 960  # 车牌的宽度self.char_interval = 18  # 字符间的间隔self.point_size = 62  # 第2个字符与第三个字符间有一个点，该点的尺寸else:raise ValueError('目前不支持该类型车牌！')def generate_images(self, plate_num_str_list):if self.plate_type in ['single_blue', 'single_yellow', ]:plate_images = self.generate_440_140_plate(plate_num_str_list)elif self.plate_type == 'small_new_energy':plate_images = self.generate_480_140_plate(plate_num_str_list)else:raise ValueError('该类型车牌目前功能尚未完成！')return plate_imagesdef generate_440_140_plate(self, plate_num_str_list):""" 生成440 * 140尺寸的7位车牌字符图片:param plate_nums::return:"""plate_images = list()for plate_num in plate_num_str_list:# 创建空白车牌号图片img = np.array(Image.new("RGB", (self.plate_width, self.plate_height), self.bg_color))# 每个字符的x轴起始、终止位置char_width_start = self.left_offsetchar_width_end = char_width_start + self.char_widthimg[:, char_width_start:char_width_end] = self.generate_char_image(plate_num[0])  # 生成的图片和img的第一维大小相同，所以在img中直接使用符号":"char_width_start = char_width_end + self.char_intervalchar_width_end = char_width_start + self.char_widthimg[:, char_width_start:char_width_end] = self.generate_char_image(plate_num[1])# 隔开特殊间隙，继续添加车牌的后续车牌号char_width_end = char_width_end + self.point_size + self.char_intervalfor i in range(2, len(plate_num)):char_width_start = char_width_end + self.char_intervalchar_width_end = char_width_start + self.char_widthimg[:, char_width_start:char_width_end] = self.generate_char_image(plate_num[i])plate_images.append(img)# chars_image debug# cv2.imshow("chars_image debug", img)# cv2.waitKey()return plate_imagesdef generate_char_image(self, char):""" 生成字符图片:param char: 字符:return:"""# 根据是否中文字符，选择生成模式if char in CharsImageGenerator.numerals or char in CharsImageGenerator.alphabet:img = self.generate_en_char_image(char)else:img = self.generate_ch_char_image(char)return imgdef generate_ch_char_image(self, char):""" 生成中文字符图片:param char: 待生成的中文字符"""img = Image.new("RGB", (self.chinese_original_width, self.plate_height), self.bg_color)ImageDraw.Draw(img).text((0, self.height_offset), char, self.fg_color, font=self.font_ch)img = img.resize((self.char_width, self.plate_height))return np.array(img)def generate_en_char_image(self, char):"""" 生成英文字符图片:param char: 待生成的英文字符"""img = Image.new("RGB", (self.english_original_width, self.plate_height), self.bg_color)ImageDraw.Draw(img).text((0, self.height_offset), char, self.fg_color, font=self.font_en)img = img.resize((self.char_width, self.plate_height))return np.array(img)class LicensePlateImageGenerator(object):"""根据车牌类型生成底牌图片"""single_blue_plate_bg = './images/single_blue.bmp'small_new_energy_plate_bg = './images/small_new_energy.jpg'def __init__(self, plate_type):self.plate_type = plate_typeif plate_type == 'single_blue':plate_image = cv2.imread(LicensePlateImageGenerator.single_blue_plate_bg)elif plate_type == 'small_new_energy':plate_image = cv2.imread(LicensePlateImageGenerator.small_new_energy_plate_bg)else:raise ValueError('该类型车牌目前功能尚未完成！')# template_image debug# cv2.imshow("template_image debug", plate_image)# cv2.waitKey()self.bg = plate_imagedef generate_template_image(self, width, height):return cv2.resize(self.bg, (width, height))class ImageAugmentation(object):"""图像增强操作：HSV变化，添加背景，高斯噪声，污渍"""horizontal_sight_directions = ('left', 'mid', 'right')vertical_sight_directions = ('up', 'mid', 'down')def __init__(self, plate_type, template_image):self.plate_type = plate_type# 确定字符颜色是否应该为黑色if plate_type == 'single_blue':# 字符为白色self.is_black_char = Falseelif plate_type in ['single_yellow', 'small_new_energy']:# 字符为黑字self.is_black_char = Trueelse:raise ValueError('暂时不支持该类型车牌')self.template_image = template_image# 透视变换self.angle_horizontal = 15self.angle_vertical = 15self.angle_up_down = 10self.angle_left_right = 5self.factor = 10# 色调，饱和度，亮度self.hue_keep = 0.8self.saturation_keep = 0.3self.value_keep = 0.2# 自然环境照片的路径列表self.env_data_paths = ImageAugmentation.search_file("background")# 高斯噪声levelself.level = 1 + ImageAugmentation.rand_reduce(4)# 污渍self.smu = cv2.imread("images/smu.jpg")def left_right_transfer(self, img, is_left=True, angle=None):"""左右视角，默认左视角:param img::param is_left::param angle: 角度:return:"""if angle is None:angle = self.angle_left_rightshape = img.shapesize_src = (shape[1], shape[0])  # width, height# 源图像四个顶点坐标pts1 = np.float32([[0, 0], [0, size_src[1]], [size_src[0], 0], [size_src[0], size_src[1]]])# 计算图片进行投影倾斜后的位置interval = abs(int(math.sin((float(angle) / 180) * math.pi) * shape[0]))# 目标图像上四个顶点的坐标if is_left:pts2 = np.float32([[0, 0], [0, size_src[1]],[size_src[0], interval], [size_src[0], size_src[1] - interval]])else:pts2 = np.float32([[0, interval], [0, size_src[1] - interval],[size_src[0], 0], [size_src[0], size_src[1]]])# 获取 3x3的投影映射/透视变换 矩阵matrix = cv2.getPerspectiveTransform(pts1, pts2)dst = cv2.warpPerspective(img, matrix, size_src)return dst, matrix, size_srcdef up_down_transfer(self, img, is_down=True, angle=None):""" 上下视角，默认下视角:param img: 正面视角原始图片:param is_down: 是否下视角:param angle: 角度:return:"""if angle is None:angle = self.rand_reduce(self.angle_up_down)shape = img.shapesize_src = (shape[1], shape[0])# 源图像四个顶点坐标pts1 = np.float32([[0, 0], [0, size_src[1]], [size_src[0], 0], [size_src[0], size_src[1]]])# 计算图片进行投影倾斜后的位置interval = abs(int(math.sin((float(angle) / 180) * math.pi) * shape[0]))# 目标图像上四个顶点的坐标if is_down:pts2 = np.float32([[interval, 0], [0, size_src[1]],[size_src[0] - interval, 0], [size_src[0], size_src[1]]])else:pts2 = np.float32([[0, 0], [interval, size_src[1]],[size_src[0], 0], [size_src[0] - interval, size_src[1]]])# 获取 3x3的投影映射/透视变换 矩阵matrix = cv2.getPerspectiveTransform(pts1, pts2)dst = cv2.warpPerspective(img, matrix, size_src)return dst, matrix, size_srcdef vertical_tilt_transfer(self, img, is_left_high=True):""" 添加按照指定角度进行垂直倾斜(上倾斜或下倾斜，最大倾斜角度self.angle_vertical一半）:param img: 输入图像的numpy:param is_left_high: 图片投影的倾斜角度，左边是否相对右边高"""angle = self.rand_reduce(self.angle_vertical)shape = img.shapesize_src = [shape[1], shape[0]]# 源图像四个顶点坐标pts1 = np.float32([[0, 0], [0, size_src[1]], [size_src[0], 0], [size_src[0], size_src[1]]])# 计算图片进行上下倾斜后的距离，及形状interval = abs(int(math.sin((float(angle) / 180) * math.pi) * shape[1]))size_target = (int(math.cos((float(angle) / 180) * math.pi) * shape[1]), shape[0] + interval)# 目标图像上四个顶点的坐标if is_left_high:pts2 = np.float32([[0, 0], [0, size_target[1] - interval],[size_target[0], interval], [size_target[0], size_target[1]]])else:pts2 = np.float32([[0, interval], [0, size_target[1]],[size_target[0], 0], [size_target[0], size_target[1] - interval]])# 获取 3x3的投影映射/透视变换 矩阵matrix = cv2.getPerspectiveTransform(pts1, pts2)dst = cv2.warpPerspective(img, matrix, size_target)return dst, matrix, size_targetdef horizontal_tilt_transfer(self, img, is_right_tilt=True):""" 添加按照指定角度进行水平倾斜(右倾斜或左倾斜，最大倾斜角度self.angle_horizontal一半）:param img: 输入图像的numpy:param is_right_tilt: 图片投影的倾斜方向（右倾，左倾）"""angle = self.rand_reduce(self.angle_horizontal)shape = img.shapesize_src = [shape[1], shape[0]]# 源图像四个顶点坐标pts1 = np.float32([[0, 0], [0, size_src[1]], [size_src[0], 0], [size_src[0], size_src[1]]])# 计算图片进行左右倾斜后的距离，及形状interval = abs(int(math.sin((float(angle) / 180) * math.pi) * shape[0]))size_target = (shape[1] + interval, int(math.cos((float(angle) / 180) * math.pi) * shape[0]))# 目标图像上四个顶点的坐标if is_right_tilt:pts2 = np.float32([[interval, 0], [0, size_target[1]],[size_target[0], 0], [size_target[0] - interval, size_target[1]]])else:pts2 = np.float32([[0, 0], [interval, size_target[1]],[size_target[0] - interval, 0], [size_target[0], size_target[1]]])# 获取 3x3的投影映射/透视变换 矩阵matrix = cv2.getPerspectiveTransform(pts1, pts2)dst = cv2.warpPerspective(img, matrix, size_target)return dst, matrix, size_targetdef sight_transfer(self, images, horizontal_sight_direction, vertical_sight_direction):"""对图片进行视角变换:param images::param horizontal_sight_direction: 水平视角变换方向:param vertical_sight_direction: 垂直视角变换方向:return:"""flag = 0img_num = len(images)# 左右视角if horizontal_sight_direction == 'left':flag += 1images[0], matrix, size = self.left_right_transfer(images[0], is_left=True)for i in range(1, img_num):images[i] = cv2.warpPerspective(images[i], matrix, size)elif horizontal_sight_direction == 'right':flag -= 1images[0], matrix, size = self.left_right_transfer(images[0], is_left=False)for i in range(1, img_num):images[i] = cv2.warpPerspective(images[i], matrix, size)else:pass# 上下视角if vertical_sight_direction == 'down':flag += 1images[0], matrix, size = self.up_down_transfer(images[0], is_down=True)for i in range(1, img_num):images[i] = cv2.warpPerspective(images[i], matrix, size)elif vertical_sight_direction == 'up':flag -= 1images[0], matrix, size = self.up_down_transfer(images[0], is_down=False)for i in range(1, img_num):images[i] = cv2.warpPerspective(images[i], matrix, size)else:pass# 左下视角 或 右上视角if abs(flag) == 2:images[0], matrix, size = self.vertical_tilt_transfer(images[0], is_left_high=True)for i in range(1, img_num):images[i] = cv2.warpPerspective(images[i], matrix, size)images[0], matrix, size = self.horizontal_tilt_transfer(images[0], is_right_tilt=True)for i in range(1, img_num):images[i] = cv2.warpPerspective(images[i], matrix, size)# 左上视角 或 右下视角elif abs(flag) == 1:images[0], matrix, size = self.vertical_tilt_transfer(images[0], is_left_high=False)for i in range(1, img_num):images[i] = cv2.warpPerspective(images[i], matrix, size)images[0], matrix, size = self.horizontal_tilt_transfer(images[0], is_right_tilt=False)for i in range(1, img_num):images[i] = cv2.warpPerspective(images[i], matrix, size)else:passreturn images@staticmethoddef search_file(search_path, file_format='.jpg'):"""在指定目录search_path下，递归目录搜索指定尾缀的文件"""file_path_list = []for root_path, dir_names, file_names in os.walk(search_path):for filename in file_names:if filename.endswith(file_format):file_path_list.append(os.path.join(root_path, filename))return file_path_list@staticmethoddef rand_reduce(val):return int(np.random.random() * val)def add_gauss(self, img, level=None):""" 添加高斯模糊:param img: 待加噪图片:param level: 加噪水平"""if level is None:level = self.levelreturn cv2.blur(img, (level * 2 + 1, level * 2 + 1))def add_single_channel_noise(self, single):""" 添加高斯噪声:param single: 单一通道的图像数据"""diff = 255 - single.max()noise = np.random.normal(0, 1 + self.rand_reduce(6), single.shape)noise = (noise - noise.min()) / (noise.max() - noise.min())noise = diff * noisenoise = noise.astype(np.uint8)dst = single + noisereturn dstdef add_noise(self, img):"""添加噪声"""img[:, :, 0] = self.add_single_channel_noise(img[:, :, 0])img[:, :, 1] = self.add_single_channel_noise(img[:, :, 1])img[:, :, 2] = self.add_single_channel_noise(img[:, :, 2])return imgdef add_smudge(self, img, smu=None):"""添加污渍"""if smu is None:smu = self.smu# 截取某一部分rows = self.rand_reduce(smu.shape[0] - img.shape[0])cols = self.rand_reduce(smu.shape[1] - img.shape[1])add_smu = smu[rows:rows + img.shape[0], cols:cols + img.shape[1]]img = cv2.bitwise_not(img)img = cv2.bitwise_and(add_smu, img)img = cv2.bitwise_not(img)return imgdef rand_environment(self, img, env_data_paths=None):""" 添加自然环境的噪声:param img: 待加噪图片:param env_data_paths: 自然环境图片路径列表"""if env_data_paths is None:env_data_paths = self.env_data_paths# 随机选取环境照片index = self.rand_reduce(len(env_data_paths))env = cv2.imread(env_data_paths[index])env = cv2.resize(env, (img.shape[1], img.shape[0]))# 找到黑背景，反转为白bak = (img == 0)for i in range(bak.shape[2]):bak[:, :, 0] &= bak[:, :, i]for i in range(bak.shape[2]):bak[:, :, i] = bak[:, :, 0]bak = bak.astype(np.uint8) * 255# 环境照片用白掩码裁剪，然后与原图非黑部分合并inv = cv2.bitwise_and(bak, env)img = cv2.bitwise_or(inv, img)return imgdef rand_hsv(self, img):""" 添加饱和度光照的噪声:param img: BGR格式的图片:return 加了饱和度、光照噪声的BGR图片"""hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)# 色调，饱和度，亮度hsv[:, :, 0] = hsv[:, :, 0] * (self.hue_keep + np.random.random() * (1 - self.hue_keep))hsv[:, :, 1] = hsv[:, :, 1] * (self.saturation_keep + np.random.random() * (1 - self.saturation_keep))hsv[:, :, 2] = hsv[:, :, 2] * (self.value_keep + np.random.random() * (1 - self.value_keep))img = cv2.cvtColor(hsv, cv2.COLOR_HSV2BGR)return imgdef augment(self, img, horizontal_sight_direction=None, vertical_sight_direction=None):"""综合上面的加载操作，进行全流程加噪:param img::param horizontal_sight_direction: 水平视角方向:param vertical_sight_direction: 垂直视角方向:return:"""if horizontal_sight_direction is None:horizontal_sight_direction = ImageAugmentation.horizontal_sight_directions[random.randint(0, 2)]if vertical_sight_direction is None:vertical_sight_direction = ImageAugmentation.vertical_sight_directions[random.randint(0, 2)]if not self.is_black_char:# 转为黑底白字img = cv2.bitwise_not(img)img = cv2.bitwise_or(img, self.template_image)# 基于视角的变换img = self.sight_transfer([img], horizontal_sight_direction, vertical_sight_direction)img = img[0]img = self.rand_environment(img)img = self.rand_hsv(img)else:# 底牌加车牌文字img = cv2.bitwise_and(img, self.template_image)# 基于视角的变换img = self.sight_transfer([img], horizontal_sight_direction, vertical_sight_direction)img = img[0]img = self.rand_environment(img)img = self.rand_hsv(img)img = self.add_gauss(img)img = self.add_noise(img)img = self.add_smudge(img)return imgclass LicensePlateGenerator(object):@staticmethoddef generate_license_plate_images(plate_type, plate_num_str_list, save_path):"""生成特定类型的的车牌图片，并保存到指定目录下:param plate_type: 车牌类型:param plate_num_str_list: 车牌号码列表:param save_path: 文件根目录:return:"""save_path = os.path.join(save_path, plate_type)if not os.path.exists(save_path):os.makedirs(save_path)print('\r>> 生成车牌号图片...')# 生成车牌号码，白底黑字chars_image_generator = CharsImageGenerator(plate_type)chars_images = chars_image_generator.generate_images(plate_num_str_list)# 生成车牌底牌license_template_generator = LicensePlateImageGenerator(plate_type)template_image = license_template_generator.generate_template_image(chars_image_generator.plate_width, chars_image_generator.plate_height)print('\r>> 生成车牌图片...')# 数据增强及车牌字符颜色修正，并保存augmentation = ImageAugmentation(plate_type, template_image)plate_height = 72plate_width = int(chars_image_generator.plate_width * plate_height / chars_image_generator.plate_height)i = 1for index, char_image in enumerate(chars_images):image_name = str(i) + "_" + plate_num_str_list[index] + ".jpg"image_path = os.path.join(save_path, image_name)image = augmentation.augment(char_image)image = cv2.resize(image, (plate_width, plate_height))cv2.imencode('.jpg', image)[1].tofile(image_path)print("\r>> {} done...".format(image_name))i += 1if __name__ == '__main__':# 保存文件夹名称file_path = os.path.join(os.getcwd(), 'plate_images')# 车牌号码列表plate_num_str_list = ["浙A5B5T3"]LicensePlateGenerator.generate_license_plate_images('single_blue', plate_num_str_list, file_path)

python+opencv生成较真实的车牌号码图片相关推荐

【目标检测】在图像上画bounding box框，生成带真实标签gt的图片
[目标检测]在图像上画bounding box框,生成带真实标签gt的图片问题/Motivation 数据格式用到的库实际代码` 结果展示问题/Motivation 在制作完数据集后,想看一下 ...
使用Python,OpenCV生成Aruco标记
这篇博客将介绍Aruco标记是什么,以及如何使用Python,OpenCV生成Aruco标记.本文提供俩种生成方式:在线生成和代码生成: 使用OpenCV处理ArUco标签非常简单,因为OpenCV库 ...
python opencv 等比例调整（缩放）图片分辨率大小代码 cv2.resize()
# -*- coding: utf-8 -*- """ @File : 200113_等比例调整图像分辨率大小.py @Time : 2020/1/13 13:38 @A ...
python opencv Intel Realsense 使用滑动条调整图片的曝光率
# -*- coding: utf-8 -*- """ @File : 200113:测试跟踪条tracebackbar().py @Time : 2020/1/13 9 ...
python opencv生成tf模型_基于TensorFlow+ Opencv 的人脸识别和模型训练
一.准备工作本次实例的anaconda 环境 (有需要的自己导入anaconda) 链接:https://pan.baidu.com/s/1IVt2ap-NYdg64uHSh-viaA 提取码:g7 ...
python opencv生成模糊图像
去除由于对焦,运动等造成的模糊图像,所以在构建数据集的时候考虑用opencv对清晰的图片进行处理获得模糊的图片从而进行训练. 1) 运动模糊图像一般来说,运动模糊的图像都是朝同一方向运动的,那么就可 ...
python opencv生成 html5 支持的mp4
c++操作mp文件: c++OpenCV操作mp4_jacke121的专栏-CSDN博客 opencv4.0 opencv3.3也支持 # -*- coding: utf-8 -*- import o ...
python opencv生成钢琴键与五线谱的对照图
代码做一个壁纸帮助学习钢琴. import cv2 import numpy as npbadu = 5 baijiankuan = 46 heijiankuan = 24 baijianchang ...
python opencv imwrite()方法 cv :: ImwriteFlags （写入图片、保存图片）
from init.py: opencv版本4.1.0,不同版本也许doc稍有不同 def imwrite(filename, img, params=None): # real signature ...

python+opencv生成较真实的车牌号码图片

python+opencv生成较真实的车牌号码图片相关推荐

最新文章

热门文章