deepellipse-circleid / code /image_precess.py
qinxunhui's picture
Add small artifacts (code, outputs, configs, LICENSE)
519ab98 verified
Raw History Blame Contribute Delete
10 kB
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
@Time : 2026/3/19 15:33
@File : image_precess.py
@Description :
"""
import os
from pathlib import Path
import cv2
import numpy as np
from matplotlib import pyplot as plt
def denoise_image(binary_image: np.ndarray) -> np.ndarray:
"""形态学开运算去噪"""
kernel = np.ones((2, 2), np.uint8)
return cv2.morphologyEx(binary_image, cv2.MORPH_OPEN, kernel)
def binarize_image(image: np.ndarray, method: str = 'adaptive') -> np.ndarray:
"""图像二值化"""
if len(image.shape) == 3:
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
else:
gray = image.copy()
if method == 'adaptive':
binary = cv2.adaptiveThreshold(
gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY_INV, 11, 2
)
elif method == 'otsu':
_, binary = cv2.threshold(
gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU
)
else:
_, binary = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY_INV)
return binary
def skeletonize_image(binary_image: np.ndarray) -> np.ndarray:
"""形态学骨架化 (迭代细化)"""
size = np.size(binary_image)
skel = np.zeros(binary_image.shape, np.uint8)
element = cv2.getStructuringElement(cv2.MORPH_CROSS, (3, 3))
done = False
temp_img = binary_image.copy()
while not done:
eroded = cv2.erode(temp_img, element)
temp = cv2.dilate(eroded, element)
temp = cv2.subtract(temp_img, temp)
skel = cv2.bitwise_or(skel, temp)
temp_img = eroded.copy()
zeros = size - cv2.countNonZero(temp_img)
if zeros == size:
done = True
return skel
def preprocess_writer_image(
image: np.ndarray,
use_binarization: bool = True,
use_skeleton: bool = True
) -> np.ndarray:
"""
完整的图像预处理流程
Args:
image: 输入图像 (RGB 或灰度)
use_binarization: 是否二值化
use_skeleton: 是否骨架化
debug_path: 调试输出路径
Returns:
处理后的图像 (单通道或三通道)
"""
# 二值化
if use_binarization:
processed = binarize_image(image, method='otsu')
else:
processed = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY) if len(image.shape) == 3 else image.copy()
# 去噪
processed = denoise_image(processed)
# 骨架化
if use_skeleton:
processed = skeletonize_image(processed)
return processed
def estimate_stroke_width_simple(binary_image, skeleton_image, k=1.5):
"""
【极简版】通过面积比估算平均笔画宽度
公式:Width ≈ Area_Binary / Area_Skeleton
"""
# 1. 统计前景像素数量 (面积)
# 假设输入是 0(背景) 和 255(前景)
binary_area = cv2.countNonZero(binary_image)
skeleton_area = cv2.countNonZero(skeleton_image)
# 2. 防止除以零
if skeleton_area == 0:
return 0.0
# 3. 计算宽度
avg_width = binary_area / skeleton_area * k
return avg_width
def estimate_stroke_width(image_orig, k=1.5):
"""
【极简版】通过面积比估算平均笔画宽度
公式:Width ≈ Area_Binary / Area_Skeleton
"""
binary = binarize_image(image_orig, method='otsu')
# B. 骨架化
skel = skeletonize_image(binary)
# C. 计算宽度
width = estimate_stroke_width_simple(binary, skel, k)
return width
def resize_image(image, target_size=100):
"""
预处理单张图片:保持长宽比 resize,边缘像素填充到目标大小
"""
if image is None:
raise ValueError("警告:无法读取图片")
h, w = image.shape[:2]
# 计算缩放比例(保持长宽比)
scale = min(target_size / h, target_size / w)
new_h = int(h * scale)
new_w = int(w * scale)
# Resize 图片
image_resized = cv2.resize(image, (new_w, new_h), interpolation=cv2.INTER_AREA)
# 计算需要填充的像素量(居中填充)
pad_top = (target_size - new_h) // 2
pad_bottom = target_size - new_h - pad_top
pad_left = (target_size - new_w) // 2
pad_right = target_size - new_w - pad_left
# 使用255填充
image_padded = cv2.copyMakeBorder(
image_resized,
pad_top, pad_bottom, pad_left, pad_right,
borderType=cv2.BORDER_CONSTANT, value=(255, 255, 255),
)
return image_padded
def sta_polor():
# 配置路径
input_folder = "/path/to/circle_id/images/images/"
output_folder = "/path/to/circle_id/outputs/width_estimate2"
# 1. 创建输出目录
os.makedirs(output_folder, exist_ok=True)
print(f"✅ 输出目录已确认:{output_folder}")
# 2. 获取图片列表 (这里处理全部图片,如果想限制数量可以加 [:100])
# img_paths = list(Path(input_folder).glob("v2_*.png"))
all_paths = list(Path(input_folder).glob("*.png"))
img_paths = [p for p in all_paths if not p.name.startswith('v2_')]
# img_paths = img_paths[:100] # 调试时可取消注释
print(f"🔍 找到 {len(img_paths)} 张图片,开始处理...")
# --- 新增:用于统计的列表 ---
all_widths = []
valid_count = 0
invalid_count = 0
count = 0
for path in img_paths:
# 读取图片 (BGR)
image_orig = cv2.imread(str(path))
if image_orig is None:
print(f"⚠️ 无法读取图片:{path}")
continue
# --- 处理流程 ---
binary = binarize_image(image_orig, method='adaptive')
skel = skeletonize_image(binary)
width = estimate_stroke_width_simple(binary, skel)
# --- 数据收集 ---
if width > 0:
all_widths.append(width)
valid_count += 1
else:
invalid_count += 1
# print(f"⚠️ 无效宽度 (骨架为空): {path.name}")
# --- 可视化准备 (保持原有逻辑) ---
h, w = image_orig.shape[:2]
image_display = image_orig.copy()
text = f"W: {width:.2f}"
font = cv2.FONT_HERSHEY_SIMPLEX
font_scale = 0.6
thickness = 2
# cv2.putText(image_display, text, (5, 45), font, font_scale, (0, 255, 0), thickness)
# cv2.putText(image_display, "Original", (15, 25), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2)
#
# binary_bgr = cv2.cvtColor(binary, cv2.COLOR_GRAY2BGR)
# cv2.putText(binary_bgr, "Binary", (15, 25), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2)
#
# skel_bgr = cv2.cvtColor(skel, cv2.COLOR_GRAY2BGR)
# cv2.putText(skel_bgr, "Skeleton", (15, 25), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 255), 2)
#
# combined_image = np.hstack([image_display, binary_bgr, skel_bgr])
#
# # --- 保存结果 ---
# output_path = os.path.join(output_folder, path.name)
# cv2.imwrite(output_path, combined_image)
count += 1
if count % 50 == 0:
print(f"已处理 {count}/{len(img_paths)} 张...")
print(f"\n🎉 图片处理完成!共处理 {count} 张,有效样本 {valid_count} 张,无效 {invalid_count} 张。")
# ================= 统计与绘图 =================
if len(all_widths) > 0:
# 1. 计算统计指标
widths_arr = np.array(all_widths)
mean_w = np.mean(widths_arr)
median_w = np.median(widths_arr)
std_w = np.std(widths_arr)
min_w = np.min(widths_arr)
max_w = np.max(widths_arr)
print("\n📊 笔画宽度统计报告:")
print(f" 平均宽度 (Mean): {mean_w:.2f} px")
print(f" 中位数 (Median): {median_w:.2f} px")
print(f" 标准差 (Std): {std_w:.2f} px")
print(f" 最小值 (Min): {min_w:.2f} px")
print(f" 最大值 (Max): {max_w:.2f} px")
# 2. 绘制直方图
plt.style.use('seaborn-v0_8-whitegrid') # 使用好看的风格
fig, ax = plt.subplots(figsize=(10, 6))
# 绘制直方图
n, bins, patches = ax.hist(all_widths, bins=50, color='#4CAF50', alpha=0.7, edgecolor='black', linewidth=0.5)
# 添加统计线
ax.axvline(mean_w, color='red', linestyle='dashed', linewidth=2, label=f'Mean: {mean_w:.2f}')
ax.axvline(median_w, color='blue', linestyle='solid', linewidth=2, label=f'Median: {median_w:.2f}')
# 添加标题和标签
ax.set_title(f'Stroke Width Distribution (N={valid_count})', fontsize=14, fontweight='bold')
ax.set_xlabel('Stroke Width (pixels)', fontsize=12)
ax.set_ylabel('Frequency (Count)', fontsize=12)
ax.legend()
# 添加文本框显示详细统计
textstr = f'Mean: {mean_w:.2f}\nMedian: {median_w:.2f}\nStd: {std_w:.2f}\nMin: {min_w:.2f}\nMax: {max_w:.2f}'
props = dict(boxstyle='round', facecolor='wheat', alpha=0.5)
ax.text(0.95, 0.95, textstr, transform=ax.transAxes, fontsize=10,
verticalalignment='top', horizontalalignment='right', bbox=props)
# 保存图表
hist_plot_path = os.path.join(output_folder, "stroke_width_distribution_train.png")
plt.savefig(hist_plot_path, dpi=150, bbox_inches='tight')
print(f"\n📈 直方图已保存至:{hist_plot_path}")
# 如果在本地有图形界面,可以显示;服务器后台运行则不会弹窗
# plt.show()
plt.close()
else:
print("\n⚠️ 没有有效的宽度数据,无法绘制直方图。请检查二值化或骨架化逻辑。")
print(f"\n💡 提示:所有结果已保存至 {output_folder}")
# ================= 使用示例 =================
if __name__ == "__main__":
path = "/path/to/circle_id/images/images/39890.png"
img = cv2.imread(path)
img_new = resize_image(img)
write_path = "/path/to/circle_id/outputs/augment2/" + path.split("/")[-1]
cv2.imwrite(write_path, img_new)