两种方式识别“传统”图片验证码

# 根据图片特点，自己写的降噪算法def clean_img(img, threshold):    width, height = img.size    for j in range(height):        for i in range(width):            point = img.getpixel((i, j))            if point == 0:                for x in range(threshold):                    if j + x >= height:                        break                    else:                        if point != img.getpixel((i, j + x)):                            img.putpixel((i, j), 1)                            break    return img

# 八值法降噪def clean_img_eight(img, threshold):    width, height = img.size    arr = [[0 for col in range(width)] for row in range(height)]    arr = array(arr)    for j in range(height):        for i in range(width):            point = img.getpixel((i, j))            if point == 0:                sum = 0                for x in range(-1, 2):                    for y in range(-1, 2):                        if i + x > width - 1 or j + y > height - 1 or \                                i + x < 0 or j + y < 0:                            sum += 1                        else:                            sum += img.getpixel((i + x, j + y))                if sum >= threshold:                    arr[j, i] = 1
    for i in range(len(arr)):        for j in range(len(arr[i])):            if arr[i, j] == 1:                img.putpixel((j, i), 1)    return img

# CFS图像切割    @staticmethod    def cut_img(img, threshold, cut_width, cut_height, width_min, width_max, height_min):        charters_imgs = []        width, height = img.size        charters_pixels = []        visited_pixels = []        pixel_arr = ImgTools.get_pixel_arr(img)        for i in range(width):            for j in range(height):                pixel = img.getpixel((i, j))                if pixel == 0 and [i, j] not in visited_pixels:                    charter_pixels = Node(i, j, pixel_arr, []).traversal()                    visited_pixels.extend(charter_pixels)                    if len(charter_pixels) > threshold:                        charters_pixels.append(charter_pixels)        for i in range(len(charters_pixels)):            x_min = 0            y_min = 0            x_max = 0            y_max = 0            # 这里是为了处理没有粘连但是垂直方向有重合的字符            width, height = img.size            tmp_img = Image.new('1', (width * 2, height * 2), 255)            for j in range(len(charters_pixels[i])):                x, y = charters_pixels[i][j]                tmp_img.putpixel((x, y), 0)                if x > x_max:                    x_max = x                else:                    if x < x_min or x_min == 0:                        x_min = x                if y > y_max:                    y_max = y                else:                    if y < y_min or y_min == 0:                        y_min = y            if width_min < x_max - x_min < width_max and y_max - y_min > height_min:                # charters_imgs.append(tmp_img.crop((x_min, y_min, x_max, y_max)))                # 这里是为了将所有的图片切成一样大，便于后期的特征提取                charters_imgs.append(tmp_img.crop((x_min, y_min, x_min + cut_width, y_min + cut_height)))        return charters_imgs

class Node:    x = 0    y = 0    graph_arr = []    visited_neighbors = []
    def __init__(self, x, y, graph_arr, visited_neighbors):        self.x = x        self.y = y        self.graph_arr = graph_arr        self.visited_neighbors = visited_neighbors
    def traversal(self):        for i in range(-1, 2):            for j in range(-1, 2):                p = self.x + i                q = self.y + j                if (0 <= p < len(self.graph_arr)) and (0 <= q < len(self.graph_arr[0])):                    if array(self.graph_arr)[p, q] == 0 and [p, q] not in self.visited_neighbors:                        self.visited_neighbors.append([p, q])                        next_node = Node(p, q, self.graph_arr, self.visited_neighbors)                        next_node.traversal()        return self.visited_neighbors