Expand 3D captcha into three subtypes: 3d_text, 3d_rotate, 3d_slider

Split the single "3d" captcha type into three independent expert models:
- 3d_text: 3D perspective text OCR (renamed from old "3d", CTC-based ThreeDCNN)
- 3d_rotate: rotation angle regression (new RegressionCNN, circular loss)
- 3d_slider: slider offset regression (new RegressionCNN, SmoothL1 loss)

CAPTCHA_TYPES expanded from 3 to 5 classes. Classifier samples updated
to 50000 (10000 per class). New generators, model, dataset, training
utilities, and full pipeline/export/CLI support for all subtypes.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Hua
2026-03-11 13:55:53 +08:00
parent 760b80ee5e
commit f5be7671bc
20 changed files with 1109 additions and 142 deletions

View File

@@ -0,0 +1,113 @@
"""
3D 滑块验证码生成器
生成滑块拼图验证码:纹理背景 + 拼图缺口 + 拼图块在左侧。
用户需将拼图块滑动到缺口位置。
标签 = 缺口 x 坐标偏移(整数)
文件名格式: {offset}_{index:06d}.png
"""
import random
from PIL import Image, ImageDraw, ImageFilter
from config import GENERATE_CONFIG
from generators.base import BaseCaptchaGenerator
class ThreeDSliderGenerator(BaseCaptchaGenerator):
"""3D 滑块验证码生成器。"""
def __init__(self, seed: int | None = None):
from config import RANDOM_SEED
super().__init__(seed=seed if seed is not None else RANDOM_SEED)
self.cfg = GENERATE_CONFIG["3d_slider"]
self.width, self.height = self.cfg["image_size"]
def generate(self, text: str | None = None) -> tuple[Image.Image, str]:
rng = self.rng
pw, ph = self.cfg["puzzle_size"]
gap_x_lo, gap_x_hi = self.cfg["gap_x_range"]
# 缺口位置
gap_x = rng.randint(gap_x_lo, gap_x_hi)
gap_y = rng.randint(10, self.height - ph - 10)
if text is None:
text = str(gap_x)
# 1. 生成纹理背景
img = self._textured_background(rng)
# 2. 从缺口位置截取拼图块内容
piece_content = img.crop((gap_x, gap_y, gap_x + pw, gap_y + ph)).copy()
# 3. 绘制缺口 (半透明灰色区域)
overlay = Image.new("RGBA", img.size, (0, 0, 0, 0))
overlay_draw = ImageDraw.Draw(overlay)
overlay_draw.rectangle(
[gap_x, gap_y, gap_x + pw, gap_y + ph],
fill=(80, 80, 80, 160),
outline=(60, 60, 60, 200),
width=2,
)
img = img.convert("RGBA")
img = Image.alpha_composite(img, overlay)
img = img.convert("RGB")
# 4. 绘制拼图块在左侧
piece_x = self.cfg["piece_left_margin"]
piece_img = Image.new("RGBA", (pw + 4, ph + 4), (0, 0, 0, 0))
piece_draw = ImageDraw.Draw(piece_img)
# 阴影
piece_draw.rectangle([2, 2, pw + 3, ph + 3], fill=(0, 0, 0, 80))
# 内容
piece_img.paste(piece_content, (0, 0))
# 边框
piece_draw.rectangle([0, 0, pw - 1, ph - 1], outline=(255, 255, 255, 200), width=2)
img_rgba = img.convert("RGBA")
img_rgba.paste(piece_img, (piece_x, gap_y), piece_img)
img = img_rgba.convert("RGB")
# 5. 轻微模糊
img = img.filter(ImageFilter.GaussianBlur(radius=0.3))
return img, text
def _textured_background(self, rng: random.Random) -> Image.Image:
"""生成带纹理的彩色背景。"""
img = Image.new("RGB", (self.width, self.height))
draw = ImageDraw.Draw(img)
# 渐变底色
base_r, base_g, base_b = rng.randint(100, 180), rng.randint(100, 180), rng.randint(100, 180)
for y in range(self.height):
ratio = y / max(self.height - 1, 1)
r = int(base_r + 30 * ratio)
g = int(base_g - 20 * ratio)
b = int(base_b + 10 * ratio)
draw.line([(0, y), (self.width, y)], fill=(r, g, b))
# 添加纹理噪声
noise_intensity = self.cfg["bg_noise_intensity"]
for _ in range(self.width * self.height // 8):
x = rng.randint(0, self.width - 1)
y = rng.randint(0, self.height - 1)
pixel = img.getpixel((x, y))
noise = tuple(
max(0, min(255, c + rng.randint(-noise_intensity, noise_intensity)))
for c in pixel
)
draw.point((x, y), fill=noise)
# 随机色块 (模拟图案)
for _ in range(rng.randint(3, 6)):
x1, y1 = rng.randint(0, self.width - 30), rng.randint(0, self.height - 20)
x2, y2 = x1 + rng.randint(15, 40), y1 + rng.randint(10, 25)
color = tuple(rng.randint(60, 220) for _ in range(3))
draw.rectangle([x1, y1, x2, y2], fill=color)
return img