211 lines
7.2 KiB
Python
211 lines
7.2 KiB
Python
"""
|
||
将3x3网格图片切分为9张独立图片
|
||
基于行/列像素方差检测分隔线位置
|
||
"""
|
||
|
||
import os
|
||
from pathlib import Path
|
||
import numpy as np
|
||
from PIL import Image
|
||
|
||
|
||
def find_split_line(variance: np.ndarray, start: int, end: int) -> tuple[int, int]:
|
||
"""
|
||
在指定区间内找到方差最小的连续区域（分隔线位置）
|
||
返回 (line_start, line_end)
|
||
"""
|
||
# 在区间内找最小值位置
|
||
region = variance[start:end]
|
||
min_idx = start + np.argmin(region)
|
||
min_val = variance[min_idx]
|
||
|
||
# 向两侧扩展，直到方差明显上升（超过最小值的3倍或绝对阈值）
|
||
threshold = max(min_val * 3, 50)
|
||
|
||
line_start = min_idx
|
||
while line_start > 0 and variance[line_start - 1] < threshold:
|
||
line_start -= 1
|
||
|
||
line_end = min_idx
|
||
while line_end < len(variance) - 1 and variance[line_end + 1] < threshold:
|
||
line_end += 1
|
||
|
||
return line_start, line_end + 1 # 返回左闭右开区间
|
||
|
||
|
||
def find_split_line_by_gradient(gray: np.ndarray, axis: int, start: int, end: int, is_bright_line: bool) -> tuple[int, int]:
|
||
"""
|
||
基于梯度突变检测分隔线边界
|
||
axis: 0=检测水平线, 1=检测垂直线
|
||
is_bright_line: True=亮色分隔线(如白色), False=暗色分隔线(如黑色)
|
||
"""
|
||
if axis == 0:
|
||
means = np.mean(gray, axis=1) # 每行的平均亮度
|
||
else:
|
||
means = np.mean(gray, axis=0) # 每列的平均亮度
|
||
|
||
# 计算相邻行/列的亮度差异（梯度）
|
||
gradient = np.diff(means)
|
||
|
||
# 找分隔线中心：在区间内找亮度极值
|
||
region = means[start:end]
|
||
if is_bright_line:
|
||
center_idx = start + np.argmax(region) # 亮色线找最亮点
|
||
else:
|
||
center_idx = start + np.argmin(region) # 暗色线找最暗点
|
||
|
||
# 从中心向左找边界：寻找梯度符号变化点（亮度突变）
|
||
line_start = center_idx
|
||
for i in range(center_idx - 1, max(0, center_idx - 30), -1):
|
||
# 对于亮色线：边界处 gradient > 0（从暗到亮）
|
||
# 对于暗色线：边界处 gradient < 0（从亮到暗）
|
||
if is_bright_line and gradient[i] > 3:
|
||
line_start = i + 1
|
||
break
|
||
elif not is_bright_line and gradient[i] < -3:
|
||
line_start = i + 1
|
||
break
|
||
line_start = i
|
||
|
||
# 从中心向右找边界
|
||
line_end = center_idx
|
||
for i in range(center_idx, min(len(gradient), center_idx + 30)):
|
||
if is_bright_line and gradient[i] < -3:
|
||
line_end = i + 1
|
||
break
|
||
elif not is_bright_line and gradient[i] > 3:
|
||
line_end = i + 1
|
||
break
|
||
line_end = i + 1
|
||
|
||
return line_start, line_end
|
||
|
||
|
||
def detect_split_lines_forest(image: Image.Image) -> tuple[list, list]:
|
||
"""专门处理 forest 的白色宽分隔线"""
|
||
gray = np.array(image.convert('L'), dtype=np.float32)
|
||
h, w = gray.shape
|
||
|
||
h_line1 = find_split_line_by_gradient(gray, 0, h // 4, h // 2 + h // 8, is_bright_line=True)
|
||
h_line2 = find_split_line_by_gradient(gray, 0, h // 2 + h // 8, 3 * h // 4, is_bright_line=True)
|
||
|
||
v_line1 = find_split_line_by_gradient(gray, 1, w // 4, w // 2 + w // 8, is_bright_line=True)
|
||
v_line2 = find_split_line_by_gradient(gray, 1, w // 2 + w // 8, 3 * w // 4, is_bright_line=True)
|
||
|
||
return [h_line1, h_line2], [v_line1, v_line2]
|
||
|
||
|
||
def detect_split_lines_snow_mountain(image: Image.Image) -> tuple[list, list]:
|
||
"""专门处理 snow_mountain 的黑色细分隔线"""
|
||
gray = np.array(image.convert('L'), dtype=np.float32)
|
||
h, w = gray.shape
|
||
|
||
h_line1 = find_split_line_by_gradient(gray, 0, h // 4, h // 2 + h // 8, is_bright_line=False)
|
||
h_line2 = find_split_line_by_gradient(gray, 0, h // 2 + h // 8, 3 * h // 4, is_bright_line=False)
|
||
|
||
v_line1 = find_split_line_by_gradient(gray, 1, w // 4, w // 2 + w // 8, is_bright_line=False)
|
||
v_line2 = find_split_line_by_gradient(gray, 1, w // 2 + w // 8, 3 * w // 4, is_bright_line=False)
|
||
|
||
return [h_line1, h_line2], [v_line1, v_line2]
|
||
|
||
|
||
def detect_split_lines(image: Image.Image) -> tuple[list, list]:
|
||
"""
|
||
检测水平和垂直分隔线的位置
|
||
返回 (h_lines, v_lines)，每个是 [(start1, end1), (start2, end2)]
|
||
"""
|
||
gray = np.array(image.convert('L'), dtype=np.float32)
|
||
h, w = gray.shape
|
||
|
||
# 计算每行的方差
|
||
row_variance = np.var(gray, axis=1)
|
||
|
||
# 计算每列的方差
|
||
col_variance = np.var(gray, axis=0)
|
||
|
||
# 平滑处理，减少噪点影响
|
||
kernel_size = 3
|
||
row_variance = np.convolve(row_variance, np.ones(kernel_size)/kernel_size, mode='same')
|
||
col_variance = np.convolve(col_variance, np.ones(kernel_size)/kernel_size, mode='same')
|
||
|
||
# 在约 1/4~1/2 区间找第一条分隔线，1/2~3/4 区间找第二条
|
||
h_line1 = find_split_line(row_variance, h // 4, h // 2 + h // 8)
|
||
h_line2 = find_split_line(row_variance, h // 2 + h // 8, 3 * h // 4)
|
||
|
||
v_line1 = find_split_line(col_variance, w // 4, w // 2 + w // 8)
|
||
v_line2 = find_split_line(col_variance, w // 2 + w // 8, 3 * w // 4)
|
||
|
||
return [h_line1, h_line2], [v_line1, v_line2]
|
||
|
||
|
||
def split_image(image: Image.Image, h_lines: list, v_lines: list) -> list[Image.Image]:
|
||
"""
|
||
根据分隔线位置切分图片为9块
|
||
"""
|
||
w, h = image.size
|
||
|
||
# 计算切分边界：[0, line1_start, line1_end, line2_start, line2_end, total]
|
||
y_bounds = [0, h_lines[0][0], h_lines[0][1], h_lines[1][0], h_lines[1][1], h]
|
||
x_bounds = [0, v_lines[0][0], v_lines[0][1], v_lines[1][0], v_lines[1][1], w]
|
||
|
||
tiles = []
|
||
# 取索引 0, 2, 4 对应的区域（跳过分隔线区域 1, 3）
|
||
for row_idx in [0, 2, 4]:
|
||
for col_idx in [0, 2, 4]:
|
||
left = x_bounds[col_idx]
|
||
right = x_bounds[col_idx + 1]
|
||
top = y_bounds[row_idx]
|
||
bottom = y_bounds[row_idx + 1]
|
||
|
||
tile = image.crop((left, top, right, bottom))
|
||
tiles.append(tile)
|
||
|
||
return tiles
|
||
|
||
|
||
def process_image(input_path: Path, output_dir: Path):
|
||
"""处理单张图片"""
|
||
image = Image.open(input_path)
|
||
name = input_path.stem
|
||
|
||
# 特殊图像使用专门的检测方法
|
||
if name == 'forest':
|
||
h_lines, v_lines = detect_split_lines_forest(image)
|
||
elif name == 'snow_mountain':
|
||
h_lines, v_lines = detect_split_lines_snow_mountain(image)
|
||
else:
|
||
h_lines, v_lines = detect_split_lines(image)
|
||
|
||
print(f"{name}: 水平分隔线 {h_lines}, 垂直分隔线 {v_lines}")
|
||
|
||
# 切分
|
||
tiles = split_image(image, h_lines, v_lines)
|
||
|
||
# 保存
|
||
for i, tile in enumerate(tiles):
|
||
output_path = output_dir / f"{name}_{i}.png"
|
||
tile.save(output_path)
|
||
print(f" 保存: {output_path.name} ({tile.size[0]}x{tile.size[1]})")
|
||
|
||
|
||
def main():
|
||
script_dir = Path(__file__).parent
|
||
input_dir = script_dir / "origin2"
|
||
output_dir = script_dir / "split2"
|
||
|
||
output_dir.mkdir(exist_ok=True)
|
||
|
||
# 支持的图片格式
|
||
extensions = {'.jpg', '.jpeg', '.png'}
|
||
|
||
for path in sorted(input_dir.iterdir()):
|
||
if path.suffix.lower() in extensions:
|
||
process_image(path, output_dir)
|
||
|
||
print(f"\n完成！输出目录: {output_dir}")
|
||
|
||
|
||
if __name__ == "__main__":
|
||
main()
|
||
|