增加了元数据清除, 感知哈希干扰和颜色直方图扰动三种处理
This commit is contained in:
parent
45579c7b87
commit
d828165547
@ -9,6 +9,7 @@ import concurrent.futures
|
||||
import numpy as np
|
||||
from PIL import ImageEnhance, ImageFilter
|
||||
from .output_handler import OutputHandler
|
||||
import io
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
@ -347,6 +348,133 @@ class PosterNotesCreator:
|
||||
logger.error(traceback.format_exc())
|
||||
return None
|
||||
|
||||
def add_phash_noise(self, image: Image.Image, intensity: float = 0.05) -> Image.Image:
|
||||
"""
|
||||
添加扰动以对抗感知哈希算法(pHash)
|
||||
通过在频域添加低频扰动实现
|
||||
|
||||
Args:
|
||||
image: 输入图像
|
||||
intensity: 扰动强度(0-1)
|
||||
|
||||
Returns:
|
||||
添加扰动后的图像
|
||||
"""
|
||||
# 灰度化处理
|
||||
gray_image = image.convert('L')
|
||||
width, height = gray_image.size
|
||||
|
||||
# 确保宽高是8的倍数(DCT通常用8x8块)
|
||||
new_width = (width // 8) * 8
|
||||
new_height = (height // 8) * 8
|
||||
if new_width != width or new_height != height:
|
||||
gray_image = gray_image.resize((new_width, new_height))
|
||||
|
||||
# 转为numpy数组
|
||||
img_array = np.array(gray_image)
|
||||
|
||||
# 简化版DCT域扰动
|
||||
# 分块处理图像
|
||||
for y in range(0, new_height, 8):
|
||||
for x in range(0, new_width, 8):
|
||||
block = img_array[y:y+8, x:x+8].astype(float)
|
||||
|
||||
# 简单DCT - 对块应用频域变化
|
||||
# 这里使用简单方法模拟DCT效果
|
||||
# 真正的DCT需要使用scipy.fftpack
|
||||
avg = np.mean(block)
|
||||
# 修改低频区块(除直流分量外)
|
||||
noise_value = random.uniform(-intensity * 10, intensity * 10)
|
||||
|
||||
# 扰动左上角的低频系数(类似于DCT中的低频区域)
|
||||
block[1:3, 1:3] += noise_value
|
||||
|
||||
# 应用回原图
|
||||
img_array[y:y+8, x:x+8] = np.clip(block, 0, 255)
|
||||
|
||||
# 转回PIL图像
|
||||
modified_image = Image.fromarray(img_array.astype(np.uint8))
|
||||
|
||||
# 调整回原始大小
|
||||
if new_width != width or new_height != height:
|
||||
modified_image = modified_image.resize((width, height), Image.LANCZOS)
|
||||
|
||||
# 将修改后的灰度通道应用到原彩色图像
|
||||
if image.mode == 'RGB':
|
||||
r, g, b = image.split()
|
||||
# 混合原始图像与修改过的灰度图
|
||||
blend_factor = 0.2 # 混合强度
|
||||
r = Image.blend(r, modified_image, blend_factor)
|
||||
g = Image.blend(g, modified_image, blend_factor)
|
||||
b = Image.blend(b, modified_image, blend_factor)
|
||||
return Image.merge('RGB', (r, g, b))
|
||||
else:
|
||||
return modified_image
|
||||
|
||||
def perturb_color_histogram(self, image: Image.Image, strength: float = 0.03) -> Image.Image:
|
||||
"""
|
||||
扰动图像的颜色直方图，对抗基于颜色统计的图像匹配
|
||||
|
||||
Args:
|
||||
image: 输入图像
|
||||
strength: 扰动强度(0-1)
|
||||
|
||||
Returns:
|
||||
处理后的图像
|
||||
"""
|
||||
# 确保为RGB模式
|
||||
if image.mode != 'RGB':
|
||||
image = image.convert('RGB')
|
||||
|
||||
# 转为numpy数组
|
||||
img_array = np.array(image)
|
||||
height, width, channels = img_array.shape
|
||||
|
||||
# 对每个通道分别处理
|
||||
for channel in range(channels):
|
||||
# 计算当前通道的直方图
|
||||
hist, _ = np.histogram(img_array[:,:,channel].flatten(), bins=64, range=(0, 256))
|
||||
|
||||
# 找出主要颜色区间 (频率高的区间)
|
||||
threshold = np.percentile(hist, 70) # 取前30%的颜色块
|
||||
significant_bins = np.where(hist > threshold)[0]
|
||||
|
||||
if len(significant_bins) > 0:
|
||||
for bin_idx in significant_bins:
|
||||
# 计算当前bin对应的颜色范围
|
||||
bin_width = 256 // 64
|
||||
color_low = bin_idx * bin_width
|
||||
color_high = (bin_idx + 1) * bin_width
|
||||
|
||||
# 创建颜色范围掩码
|
||||
mask = (img_array[:,:,channel] >= color_low) & (img_array[:,:,channel] < color_high)
|
||||
|
||||
if np.any(mask):
|
||||
# 生成随机偏移值
|
||||
offset = int(strength * bin_width * (random.random() - 0.5) * 2)
|
||||
|
||||
# 应用偏移，确保在0-255范围内
|
||||
img_array[:,:,channel][mask] = np.clip(
|
||||
img_array[:,:,channel][mask] + offset, 0, 255).astype(np.uint8)
|
||||
|
||||
# 转回PIL图像
|
||||
return Image.fromarray(img_array)
|
||||
|
||||
def strip_metadata(self, image: Image.Image) -> Image.Image:
|
||||
"""
|
||||
移除图像中的所有元数据
|
||||
|
||||
Args:
|
||||
image: 输入图像
|
||||
|
||||
Returns:
|
||||
无元数据的图像
|
||||
"""
|
||||
# 创建无元数据的副本
|
||||
data = io.BytesIO()
|
||||
image.save(data, format=image.format if image.format else 'PNG')
|
||||
return Image.open(data)
|
||||
|
||||
def optimized_process_image(
|
||||
self,
|
||||
image: Image.Image,
|
||||
@ -356,7 +484,7 @@ class PosterNotesCreator:
|
||||
variation_strength: str = "medium",
|
||||
extra_effects: bool = True
|
||||
) -> Image.Image:
|
||||
"""优化后的图像处理方法，使用更高效的算法"""
|
||||
"""优化后的图像处理方法，使用更高效的算法，添加反查重技术"""
|
||||
# 设置随机种子
|
||||
if seed is not None:
|
||||
random.seed(seed)
|
||||
@ -419,12 +547,14 @@ class PosterNotesCreator:
|
||||
result = resized_image.crop((0, crop_y1, resized_width, crop_y2))
|
||||
|
||||
# 如果不需要变化或是低强度且禁用额外效果
|
||||
if not add_variation or (variation_strength == "low" and not use_extra):
|
||||
if not add_variation:
|
||||
# 重置随机种子
|
||||
if seed is not None:
|
||||
random.seed()
|
||||
np.random.seed()
|
||||
return result
|
||||
|
||||
# 清除元数据后返回
|
||||
return self.strip_metadata(result)
|
||||
|
||||
# 高效应用基本变化
|
||||
processed_image = result.convert('RGB')
|
||||
@ -450,6 +580,24 @@ class PosterNotesCreator:
|
||||
if abs(rotation_angle) > 0.1: # 只有当角度足够大时才旋转
|
||||
processed_image = processed_image.rotate(rotation_angle, resample=Image.BICUBIC, expand=False)
|
||||
|
||||
# 5. 新增 - 应用反查重技术
|
||||
# 根据变化强度选择性应用
|
||||
if use_extra:
|
||||
# 随机决定应用哪些反查重技术
|
||||
apply_phash = random.random() < 0.7
|
||||
apply_color = random.random() < 0.7
|
||||
|
||||
# 感知哈希干扰 (在中高强度变化时应用)
|
||||
if apply_phash and variation_strength != "low":
|
||||
phash_intensity = 0.05 if variation_strength == "medium" else 0.08
|
||||
processed_image = self.add_phash_noise(processed_image, phash_intensity)
|
||||
|
||||
# 颜色直方图扰动
|
||||
if apply_color:
|
||||
color_strength = 0.02 if variation_strength == "low" else \
|
||||
0.04 if variation_strength == "medium" else 0.06
|
||||
processed_image = self.perturb_color_histogram(processed_image, color_strength)
|
||||
|
||||
# 应用额外效果 (只在需要时)
|
||||
if use_extra:
|
||||
# 根据强度决定是否应用特定效果
|
||||
@ -483,6 +631,9 @@ class PosterNotesCreator:
|
||||
offset_y = random.randint(0, border_size*2)
|
||||
processed_image = bordered.crop((offset_x, offset_y, offset_x + w, offset_y + h))
|
||||
|
||||
# 6. 始终清除元数据 - 最后一步
|
||||
processed_image = self.strip_metadata(processed_image)
|
||||
|
||||
# 重置随机种子
|
||||
if seed is not None:
|
||||
random.seed()
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user