标记方式有Trimap和Strokes两种的静态图像抠图算法python源码精简版整理

最新推荐文章于 2024-09-01 08:26:16 发布

海宝7号

最新推荐文章于 2024-09-01 08:26:16 发布

阅读量2.1k

点赞数

分类专栏： python2021 文章标签： python keras tensorflow

本文链接：https://blog.csdn.net/dongbao520/article/details/121446234

版权

python2021 专栏收录该内容

186 篇文章 12 订阅

订阅专栏

import random
import math
import numpy as np
import cv2

# 标记方式有Trimap和Strokes两种的静态图像抠图算法

def gen_trimap(alpha):
    k_size = random.choice(range(1, 5))  # 随机选择核大小
    iterations = np.random.randint(1, 20)
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (k_size, k_size))
    dilated = cv2.dilate(alpha, kernel, iterations)
    eroded = cv2.erode(alpha, kernel, iterations)
    trimap = np.zeros(alpha.shape)
    trimap.fill(128)
    trimap[eroded >= 255] = 255
    trimap[dilated <= 0] = 0
    cv2.namedWindow('gaus2', cv2.WINDOW_NORMAL)  # 设置为WINDOW_NORMAL可以任意缩放
    cv2.imshow('gaus2',trimap)
    cv2.waitKey(0)
    cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/trimap.jpg", trimap)

    return trimap

def composite4(fg, bg, a, w, h):
    fg = np.array(fg, np.float32)
    bg_h, bg_w = bg.shape[:2]
    x = 0
    if bg_w > w:
        x = np.random.randint(0, bg_w - w)
    y = 0
    if bg_h > h:
        y = np.random.randint(0, bg_h - h)
    bg = np.array(bg[y:y + h, x:x + w], np.float32)
    alpha = np.zeros((h, w, 1), np.float32)
    alpha[:, :, 0] = a / 255.
    im = alpha * fg + (1 - alpha) * bg
    im = im.astype(np.uint8)
    return im, a, fg, bg

def process(fg_file, alpha_file, bg_file):
    # 备注
    # https://www.aiuai.cn/aifarm1569.html?replyTo=1690
    # 输入的图像格式不对。。。前两张一样的图片大小，第三张合成。背景合成--只能加背景图片三合一的改进。
    im = cv2.imread(fg_file)
    a = cv2.imread(alpha_file, 0)   #变成原始的灰度图

    # cv2.namedWindow('alphaa', cv2.WINDOW_NORMAL)  # 设置为WINDOW_NORMAL可以任意缩放
    # cv2.imshow('alphaa', a)
    # cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/alphaa.jpg", a)

    h, w = im.shape[:2]
    print("im_shape==", im.shape)
    # print("a=",a)
    # print("h=",h)
    # print("w=",w)
    bg = cv2.imread(bg_file)
    bh, bw = bg.shape[:2]
    print("bg_shape==", bg.shape)
    # print("bh=",bh)
    # print("bw=",bw)

    wratio = w / bw
    hratio = h / bh
    print("wratio=",wratio)
    print("hratio=",hratio)

    ratio = wratio if wratio > hratio else hratio
    if ratio > 1:
        bg = cv2.resize(src=bg, dsize=(math.ceil(bw * ratio), math.ceil(bh * ratio)), interpolation=cv2.INTER_CUBIC)

    return composite4(im, bg, a, w, h)

if __name__ == "__main__":
    #
    fg_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/37.tiff'
    alpha_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/35.tiff'
    bg_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/3.jpg'
    # fg_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/001.jpg'
    # alpha_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/002.jpg'
    # bg_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/3.jpg'
    # fg_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/006.jpg'
    # alpha_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/005.jpg'
    # bg_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/2.jpg'

    img, alpha, fg, bg = process(fg_file, alpha_file, bg_file)
    cv2.namedWindow('alpha0', cv2.WINDOW_NORMAL)  # 设置为WINDOW_NORMAL可以任意缩放
    cv2.imshow('alpha0',img)
    cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/alpha0.jpg", img)
    cv2.namedWindow('alpha1', cv2.WINDOW_NORMAL)  # 设置为WINDOW_NORMAL可以任意缩放
    cv2.imshow('alpha1',alpha)
    cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/alpha1.jpg", alpha)
    cv2.namedWindow('alpha2', cv2.WINDOW_NORMAL)  # 设置为WINDOW_NORMAL可以任意缩放
    cv2.imshow('alpha2', fg)
    cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/alpha2.jpg", fg)
    cv2.namedWindow('alpha3', cv2.WINDOW_NORMAL)  # 设置为WINDOW_NORMAL可以任意缩放
    cv2.imshow('alpha3', bg)
    cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/alpha3.jpg", bg)

    # print("alpha=",alpha)
    # print("fg=",fg)
    # print("bg=",bg)
    # print(bg)
    # print("fg==",len(fg))
    # print("bg==", len(bg))
    # print("fg_size==",fg.size)
    # print("bg_size==", bg.size)
    print("fg_shape==",fg.shape)
    print("bg_shape==", bg.shape)

    # 阿尔法通道是一个8位的灰度通道，该通道用256级灰度来记录图像中的透明度信息，定义透明、不透明和半透明区域，
    # 其中黑表示透明，白表示不透明，灰表示半透明。https://zhidao.baidu.com/question/1817884346481655668.html

    # trimap
    trimap = gen_trimap(alpha)
    #
    # alpha = xxxx
    # assert (np.unique(alpha_np) == [0, 255]).any()
    # alpha = alpha * 255
    # gen_trimap(alpha)
    #
    # 翻转
    if np.random.random_sample() > 0.5:
        img = np.fliplr(img)
        trimap = np.fliplr(trimap)
        alpha = np.fliplr(alpha)

精简版实现：

import random
import math
import numpy as np
import cv2

# 标记方式有Trimap和Strokes两种的静态图像抠图算法

def gen_trimap(alpha):
    k_size = random.choice(range(1, 5))  # 随机选择核大小
    iterations = np.random.randint(1, 20)
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (k_size, k_size))
    dilated = cv2.dilate(alpha, kernel, iterations)
    eroded = cv2.erode(alpha, kernel, iterations)
    trimap = np.zeros(alpha.shape)
    trimap.fill(128)
    trimap[eroded >= 255] = 255
    trimap[dilated <= 0] = 0
    cv2.namedWindow('gaus2', cv2.WINDOW_NORMAL)  # 设置为WINDOW_NORMAL可以任意缩放
    cv2.imshow('gaus2',trimap)
    cv2.waitKey(0)
    cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/trimap.jpg", trimap)

    return trimap

def process(fg_file, alpha_file, bg_file):

    im = cv2.imread(fg_file)
    a = cv2.imread(alpha_file, 0)   #变成原始的灰度图

    h, w = im.shape[:2]
    print("im_shape==", im.shape)

    bg = cv2.imread(bg_file)
    bh, bw = bg.shape[:2]
    print("bg_shape==", bg.shape)
    wratio = w / bw
    hratio = h / bh
    print("wratio=",wratio)
    print("hratio=",hratio)

    ratio = wratio if wratio > hratio else hratio
    if ratio > 1:
        bg = cv2.resize(src=bg, dsize=(math.ceil(bw * ratio), math.ceil(bh * ratio)), interpolation=cv2.INTER_CUBIC)

    return composite4(im, bg, a, w, h)

if __name__ == "__main__":
    fg_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/37.tiff'
    alpha_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/35.tiff'
    bg_file = 'D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/3.jpg'

    img, alpha, fg, bg = process(fg_file, alpha_file, bg_file)
    cv2.namedWindow('alpha0', cv2.WINDOW_NORMAL)  # 设置为WINDOW_NORMAL可以任意缩放
    cv2.imshow('alpha0',img)
    cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/alpha0.jpg", img)
    cv2.namedWindow('alpha1', cv2.WINDOW_NORMAL)  # 设置为WINDOW_NORMAL可以任意缩放
    cv2.imshow('alpha1',alpha)
    cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/alpha1.jpg", alpha)
    cv2.namedWindow('alpha2', cv2.WINDOW_NORMAL)  
    cv2.imshow('alpha2', fg)
    cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/alpha2.jpg", fg)
    cv2.namedWindow('alpha3', cv2.WINDOW_NORMAL) 
    cv2.imshow('alpha3', bg)
    cv2.imwrite("D:/Pictures/nbnb11/gaussian/gaussian_11tower/data01/alpha3.jpg", bg)
    print("fg_shape==",fg.shape)
    print("bg_shape==", bg.shape)

效果如图
在这里插入图片描述