针对文字图片使用pillow对图片进行操作并判断图片相似度

前言：最近想给维护的爬虫项目加点功能，摆脱人工实现自动化定时爬取，遇到的第一个难题就是登录获取cookie需要点触验证码识别，然后就进入了无止境的判断图片相似度，花了一周左右时间的尝试和查找资料，最终解决了这个问题，在此记录一下。
pillow基础：Python图像处理PIL各模块详细介绍
样例图片：

第一步：将图片处理成黑白图片
相关资料：使用python PIL库实现简单验证码的去噪

import numpy
from PIL import Image


def get_bin_table(threshold=140):
    """
    获取灰度转二值的映射table
    :param threshold:
    :return:
    """
    table = []
    for i in range(256):
        if i < threshold:
            table.append(0)
        else:
            table.append(1)

    return table


image = Image.open('1.jpg')
imgry = image.convert('L')  # 转化为灰度图
table = get_bin_table()
out = imgry.point(table, '1')

第二步：分割图片上下部分

out_up = out.crop((0, 0, out.width, out.height-100))
out_down = out.crop((1, out.height-39, out.width-200, out.height-1))

第三步：图片去噪
相关资料：使用python PIL库实现简单验证码的去噪

def sum_9_region(img, x, y, color):
    """
    9邻域框,以当前点为中心的田字框,color点个数
    :param x:
    :param y:
    :param color: 白点为1，黑点为0
    :return:
    """
    # todo 判断图片的长宽度下限
    cur_pixel = img.getpixel((x, y))  # 当前像素点的值
    width = img.width
    height = img.height

    if cur_pixel != color:  # 如果当前点为非color点区域,则不统计邻域值
        return 0

    if y == 0:  # 第一行
        if x == 0:  # 左上顶点,4邻域
            # 中心点旁边3个点
            sum = cur_pixel \
                  + img.getpixel((x, y + 1)) \
                  + img.getpixel((x + 1, y)) \
                  + img.getpixel((x + 1, y + 1))

            if color:
                return sum
            else:
                return 3-sum
        elif x == width - 1:  # 右上顶点
            sum = cur_pixel \
                  + img.getpixel((x, y + 1)) \
                  + img.getpixel((x - 1, y)) \
                  + img.getpixel((x - 1, y + 1))

            if color:
                return sum
            else:
                return 3 - sum
        else:  # 最上非顶点,6邻域
            sum = img.getpixel((x - 1, y)) \
                  + img.getpixel((x - 1, y + 1)) \
                  + cur_pixel \
                  + img.getpixel((x, y + 1)) \
                  + img.getpixel((x + 1, y)) \
                  + img.getpixel((x + 1, y + 1))
            if color:
                return sum
            else:
                return 6 - sum
    elif y == height - 1:  # 最下面一行
        if x == 0:  # 左下顶点
            # 中心点旁边3个点
            sum = cur_pixel \
                  + img.getpixel((x + 1, y)) \
                  + img.getpixel((x + 1, y - 1)) \
                  + img.getpixel((x, y - 1))
            if color:
                return sum
            else:
                return 3 - sum
        elif x == width - 1:  # 右下顶点
            sum = cur_pixel \
                  + img.getpixel((x, y - 1)) \
                  + img.getpixel((x - 1, y)) \
                  + img.getpixel((x - 1, y - 1))

            if color:
                return sum
            else:
                return 3 - sum
        else:  # 最下非顶点,6邻域
            sum = cur_pixel \
                  + img.getpixel((x - 1, y)) \
                  + img.getpixel((x + 1, y)) \
                  + img.getpixel((x, y - 1)) \
                  + img.getpixel((x - 1, y - 1)) \
                  + img.getpixel((x + 1, y - 1))
            if color:
                return sum
            else:
                return 6 - sum
    else:  # y不在边界
        if x == 0:  # 左边非顶点
            sum = img.getpixel((x, y - 1)) \
                  + cur_pixel \
                  + img.getpixel((x, y + 1)) \
                  + img.getpixel((x + 1, y - 1)) \
                  + img.getpixel((x + 1, y)) \
                  + img.getpixel((x + 1, y + 1))

            if color:
                return sum
            else:
                return 6 - sum
        elif x == width - 1:  # 右边非顶点
            # print('%s,%s' % (x, y))
            sum = img.getpixel((x, y - 1)) \
                  + cur_pixel \
                  + img.getpixel((x, y + 1)) \
                  + img.getpixel((x - 1, y - 1)) \
                  + img.getpixel((x - 1, y)) \
                  + img.getpixel((x - 1, y + 1))

            if color:
                return sum
            else:
                return 6 - sum
        else:  # 具备9领域条件的
            sum = img.getpixel((x - 1, y - 1)) \
                  + img.getpixel((x - 1, y)) \
                  + img.getpixel((x - 1, y + 1)) \
                  + img.getpixel((x, y - 1)) \
                  + cur_pixel \
                  + img.getpixel((x, y + 1)) \
                  + img.getpixel((x + 1, y - 1)) \
                  + img.getpixel((x + 1, y)) \
                  + img.getpixel((x + 1, y + 1))
            if color:
                return sum
            else:
                return 9 - sum


def collect_noise_point(img, color):
    '''收集所有的噪点'''
    noise_point_list = []
    for x in range(img.width):
        for y in range(img.height):
            res_9 = sum_9_region(img, x, y, color)
            if (0 < res_9 < 6) and img.getpixel((x, y)) == color:  # 找到孤立点
                pos = (x, y)
                noise_point_list.append(pos)
    return noise_point_list


def remove_noise_pixel(img, noise_point_list, color):
    '''根据噪点的位置信息，消除二值图片的color点噪声'''
    for item in noise_point_list:
        img.putpixel((item[0], item[1]), 1-color)


noise_point_list = collect_noise_point(out_up, 1)
# print(noise_point_list)
remove_noise_pixel(out_up, noise_point_list, 1)

第四步：分割图片

def judge(fonts_point_list, x, y, color):
    """
    判断x,y是否应该加入fonts_point_list
    :param color: 白字为70， 黑字为30
    """
    if not color:
        distance = 30
    else:
        distance = 70
    for i, j in fonts_point_list:
        if abs(x - i) < distance and abs(y - j) < distance:
            return False
    return True


def find_fonts_hang(img, color, distance):
    """
    将带有文字的图片坐标保存下来(一行一行的找)
    :param color: 1代表白色， 0代表黑色
    :param distance: 字与字之间像素的距离，白字之间为50， 黑字之间为15
    """
    fonts_point_list = []
    y = -1
    while y+distance < img.height:
        if len(fonts_point_list) == 4 or y+distance > img.height:
            break
        y += 1
        x = 0
        while x+distance < img.width:
            # print(x, y)
            res_9 = sum_9_region(img, x, y, color)
            if res_9 >= 7 and img.getpixel((x, y)) == color:  # 找到文字左上角的点
                f = False
                if not fonts_point_list:
                    f = True
                if f or judge(fonts_point_list, x, y, color):
                    pos = (x, y)
                    fonts_point_list.append(pos)
            x += 1
    print(',,', img.width, img.height)
    return fonts_point_list


def copy_fonts(img, fonts_point_list, color, distance):
    """
    将带有文字的图片部分复制下来
    :param color: 1代表白色， 0代表黑色
    :param distance: 字与字之间像素的距离，白字之间为50， 黑字之间为20
    """
    fonts_img = []
    for l_x, l_y in fonts_point_list:
        try:
            left = l_x-distance
            if left < 0:
                left = 0
            font_img = img.crop((left, l_y, l_x+distance, l_y+distance))
        except SystemError:
            print('越界')
        fonts_img.append(font_img)

    return fonts_img


fonts_point_list = find_fonts_hang(out_up, 1, 50)
fonts_img1 = copy_fonts(out_up, fonts_point_list, 1, 40)  # 获取到4张带有字的图片（白字）
fonts_img_down = find_fonts_hang(out_down, 0, 15)  
fonts_img0 = copy_fonts(out_down, fonts_img_down, 0, 20)  # 获取到4张带有字的图片（黑字）

第五步：底部照片白底黑字转换成黑底白字，将图片黑色部分切割掉

def blacktowhite(fonts_img):
    """
    将图片黑白转换
    """
    res = []
    for img in fonts_img:
        for x in range(img.width):
            for y in range(img.height):
                value = img.getpixel((x, y))
                img.putpixel((x, y), 1-value)
        # img.show()
        res.append(img)
    return res


def cut_black(img):
    """
    将图片的左右黑色部分去掉
    """
    # print(img)
    edge_left, edge_right = find_min_edge(ont_to_two(list(img.getdata()), img.width))
    img = img.crop((edge_left, 0, edge_right, img.height))
    # img.show()
    return img
fonts_img0 = blacktowhite(fonts_img0)
fonts_img0 = [cut_black(img) for img in fonts_img0]  # 下
fonts_img1 = [cut_black(img) for img in fonts_img1]  # 上

第六步：判断图片相似度
相似度判断算法：计算并探究图像原图、旋转、添加噪声的相似度

def find_min_edge_hang(data1):
    """
    找到data1主体的最小边界长度,一维数组
    """
    index_1 = 0
    index_last1 = 0
    for i in range(len(data1)):
        if data1[i] == 1:
            index_1 = i
            break
    for i in range(len(data1)):
        if data1[len(data1) - 1 - i] == 1:
            index_last1 = len(data1) - 1 - i
            break
    return index_1, index_last1


def find_min_edge(data1):
    """
    找到data1主体的最小边界长度,data1是二维数组
    """
    edge_left = []
    edge_right = []
    for hang in data1:
        index, index_last = find_min_edge_hang(hang)
        edge_left.append(index)
        edge_right.append(index_last)
    # print(min(edge_left), max(edge_right))
    return min(edge_left), max(edge_right)


def ont_to_two(data, lie):
    """
    将行存储一维数组转化为每行lie个元素的二维数组
    """
    two = []
    i = 0
    while i < len(data):
        hang = data[i: i+lie]
        i += lie
        two.append(hang)
        # print(hang)
    return two


def image_similarity_vectors_via_numpy(img1, img2):
    img1 = img1.resize((32, 32), Image.ANTIALIAS)
    img2 = img2.resize((32, 32), Image.ANTIALIAS)
    # img1.show()
    # img2.show()
    images = [img1, img2]
    vectors = []
    norms = []
    for image in images:
        vector = []
        for pixel_tuple in image.getdata():
            vector.append(numpy.average(pixel_tuple))
        vectors.append(vector)
        norms.append(numpy.linalg.norm(vector, 2))
    a, b = vectors
    a_norm, b_norm = norms
    res = numpy.dot(a / a_norm, b / b_norm)
    return res


res = []
has_select = []
similarity = {}
for img1_index in range(len(fonts_img1)):
    # similarity = {}
    for img2_index in range(len(fonts_img0)):
        similarity[(img1_index, img2_index)] = image_similarity_vectors_via_numpy(fonts_img1[img1_index], fonts_img0[img2_index])
        # break
    # break
similarity = sorted(similarity.items(), key=lambda x: x[1], reverse=True)

for _ in range(4):
    index1 = similarity[0][0][0]
    index2 = similarity[0][0][1]
    res.append(similarity[0])
    img1 = fonts_img1[index1]
    img2 = fonts_img0[index2]
    img1.show()
    img2.show()
    i = 0
    while i < len(similarity):
        if similarity[i][0][0] == index1 or similarity[i][0][1] == index2:
            similarity.remove(similarity[i])
        else:
            i += 1
    similarity = sorted(similarity, key=lambda x: x[1], reverse=True)
print(res)

配对效果图：

感谢以下文章：
Python图像处理PIL各模块详细介绍
使用python PIL库实现简单验证码的去噪
计算并探究图像原图、旋转、添加噪声的相似度