Files
douyin_guaji-main/x_template_lab.py
T
Your Name 223c127e11 chore: Clean up and refactor codebase for improved maintainability
- Removed obsolete command-line test parameters from `run_cli.py` to streamline functionality.
- Introduced a new `ocr_preprocess.py` module for unified OCR image preprocessing, enhancing recognition stability.
- Updated various components to utilize the new preprocessing function, improving code organization and readability.
- Adjusted configuration values in `config.json` for better performance and consistency.
- Added comprehensive technical documentation for the Douyin project, detailing code modification standards and operational guidelines.
- Cleaned up excessive blank lines across multiple files to enhance overall readability.
2025-12-30 23:11:34 +08:00

551 lines
22 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
x图标模板匹配阈值优化测试脚本
测试不同阈值参数对模板匹配结果的影响
"""
import sys
import os
import cv2
import numpy as np
from pathlib import Path
import glob
from collections import defaultdict
# 设置UTF-8编码
if sys.platform == 'win32':
sys.stdout.reconfigure(encoding='utf-8')
# 测试图片路径(自动扫描debug_captures文件夹)
DEBUG_CAPTURES_DIR = r"pic/debug_captures"
TEMPLATE_PATH = r"assets/x.png"
# ROI坐标(基于1080x2400分辨率)
ROI_X1, ROI_Y1 = 972, 147
ROI_X2, ROI_Y2 = 1047, 216
def load_image(image_path):
"""加载图片(支持中文路径)"""
if not os.path.exists(image_path):
raise FileNotFoundError(f"图片不存在: {image_path}")
# 使用numpy读取,支持中文路径
img_array = np.fromfile(image_path, dtype=np.uint8)
img = cv2.imdecode(img_array, cv2.IMREAD_GRAYSCALE)
if img is None:
raise ValueError(f"无法读取图片: {image_path}")
return img
def crop_roi(img, x1, y1, x2, y2):
"""裁剪ROI区域"""
h, w = img.shape[:2]
# 如果图片尺寸不是1080x2400,需要缩放坐标
if w != 1080 or h != 2400:
scale_x = w / 1080.0
scale_y = h / 2400.0
x1 = int(x1 * scale_x)
y1 = int(y1 * scale_y)
x2 = int(x2 * scale_x)
y2 = int(y2 * scale_y)
roi = img[y1:y2, x1:x2]
return roi
def test_fixed_threshold_binary(roi_img, template, threshold_values):
"""测试固定阈值二值化的不同阈值"""
results = []
for thresh in threshold_values:
try:
_, roi_bin = cv2.threshold(roi_img, thresh, 255, cv2.THRESH_BINARY)
_, tpl_bin = cv2.threshold(template, thresh, 255, cv2.THRESH_BINARY)
# 确保模板尺寸匹配
th, tw = tpl_bin.shape[:2]
rh, rw = roi_bin.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
tpl_bin = cv2.resize(tpl_bin, (max(1, int(tw * scale)), max(1, int(th * scale))))
res = cv2.matchTemplate(roi_bin, tpl_bin, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results.append(("固定阈值二值化", thresh, float(max_val or 0.0)))
except Exception as e:
results.append(("固定阈值二值化", thresh, 0.0))
return results
def test_adaptive_threshold(roi_img, template, block_sizes, c_values):
"""测试自适应二值化的不同参数"""
results = []
for block_size in block_sizes:
for c in c_values:
try:
roi_bin = cv2.adaptiveThreshold(
roi_img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, block_size, c
)
tpl_bin = cv2.adaptiveThreshold(
template, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, block_size, c
)
# 确保模板尺寸匹配
th, tw = tpl_bin.shape[:2]
rh, rw = roi_bin.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
tpl_bin = cv2.resize(tpl_bin, (max(1, int(tw * scale)), max(1, int(th * scale))))
res = cv2.matchTemplate(roi_bin, tpl_bin, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results.append(("自适应二值化", f"blockSize={block_size}, C={c}", float(max_val or 0.0)))
except Exception as e:
results.append(("自适应二值化", f"blockSize={block_size}, C={c}", 0.0))
return results
def test_canny_edges(roi_img, template, low_thresholds, high_thresholds):
"""测试Canny边缘检测的不同阈值组合"""
results = []
for low in low_thresholds:
for high in high_thresholds:
if high <= low:
continue
try:
roi_edges = cv2.Canny(roi_img, low, high)
tpl_edges = cv2.Canny(template, low, high)
# 确保模板尺寸匹配
th, tw = tpl_edges.shape[:2]
rh, rw = roi_edges.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
tpl_edges = cv2.resize(tpl_edges, (max(1, int(tw * scale)), max(1, int(th * scale))))
res = cv2.matchTemplate(roi_edges, tpl_edges, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results.append(("Canny边缘检测", f"low={low}, high={high}", float(max_val or 0.0)))
except Exception as e:
results.append(("Canny边缘检测", f"low={low}, high={high}", 0.0))
return results
def test_template_methods(roi_img, template, method_name="固定阈值二值化", thresh=180):
"""测试不同的模板匹配方法"""
# 预处理
_, roi_bin = cv2.threshold(roi_img, thresh, 255, cv2.THRESH_BINARY)
_, tpl_bin = cv2.threshold(template, thresh, 255, cv2.THRESH_BINARY)
# 确保模板尺寸匹配
th, tw = tpl_bin.shape[:2]
rh, rw = roi_bin.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
tpl_bin = cv2.resize(tpl_bin, (max(1, int(tw * scale)), max(1, int(th * scale))))
methods = [
("TM_CCOEFF_NORMED", cv2.TM_CCOEFF_NORMED),
("TM_CCORR_NORMED", cv2.TM_CCORR_NORMED),
("TM_SQDIFF_NORMED", cv2.TM_SQDIFF_NORMED),
]
results = []
for name, method in methods:
try:
res = cv2.matchTemplate(roi_bin, tpl_bin, method)
if method == cv2.TM_SQDIFF_NORMED:
_min_val, _max_val, min_loc, _max_loc = cv2.minMaxLoc(res)
score = 1.0 - float(_min_val or 0.0) # SQDIFF越小越好,转换为越大越好
else:
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
score = float(max_val or 0.0)
results.append((name, score))
except Exception as e:
results.append((name, 0.0))
return results
def test_advanced_preprocessing(roi_img, template):
"""测试高级预处理方法"""
results = []
# 1. 反转图像 + 二值化
try:
roi_inv = 255 - roi_img
tpl_inv = 255 - template
_, roi_bin_inv = cv2.threshold(roi_inv, 180, 255, cv2.THRESH_BINARY)
_, tpl_bin_inv = cv2.threshold(tpl_inv, 180, 255, cv2.THRESH_BINARY)
th, tw = tpl_bin_inv.shape[:2]
rh, rw = roi_bin_inv.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
tpl_bin_inv = cv2.resize(tpl_bin_inv, (max(1, int(tw * scale)), max(1, int(th * scale))))
res = cv2.matchTemplate(roi_bin_inv, tpl_bin_inv, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results.append(("反转+二值化", "thresh=180", float(max_val or 0.0)))
except Exception:
results.append(("反转+二值化", "thresh=180", 0.0))
# 2. 直方图均衡化
try:
roi_eq = cv2.equalizeHist(roi_img)
tpl_eq = cv2.equalizeHist(template)
th, tw = tpl_eq.shape[:2]
rh, rw = roi_eq.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
tpl_eq = cv2.resize(tpl_eq, (max(1, int(tw * scale)), max(1, int(th * scale))))
res = cv2.matchTemplate(roi_eq, tpl_eq, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results.append(("直方图均衡化", "", float(max_val or 0.0)))
except Exception:
results.append(("直方图均衡化", "", 0.0))
# 3. 对比度增强(CLAHE)
try:
clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
roi_clahe = clahe.apply(roi_img)
tpl_clahe = clahe.apply(template)
th, tw = tpl_clahe.shape[:2]
rh, rw = roi_clahe.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
tpl_clahe = cv2.resize(tpl_clahe, (max(1, int(tw * scale)), max(1, int(th * scale))))
res = cv2.matchTemplate(roi_clahe, tpl_clahe, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results.append(("CLAHE对比度增强", "clipLimit=2.0", float(max_val or 0.0)))
except Exception:
results.append(("CLAHE对比度增强", "clipLimit=2.0", 0.0))
# 4. 高斯模糊 + 边缘检测
try:
roi_blur = cv2.GaussianBlur(roi_img, (3, 3), 0)
tpl_blur = cv2.GaussianBlur(template, (3, 3), 0)
roi_edges = cv2.Canny(roi_blur, 50, 150)
tpl_edges = cv2.Canny(tpl_blur, 50, 150)
th, tw = tpl_edges.shape[:2]
rh, rw = roi_edges.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
tpl_edges = cv2.resize(tpl_edges, (max(1, int(tw * scale)), max(1, int(th * scale))))
res = cv2.matchTemplate(roi_edges, tpl_edges, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results.append(("高斯模糊+边缘检测", "blur(3,3), canny(50,150)", float(max_val or 0.0)))
except Exception:
results.append(("高斯模糊+边缘检测", "blur(3,3), canny(50,150)", 0.0))
# 5. 形态学操作(开运算)
try:
_, roi_bin = cv2.threshold(roi_img, 180, 255, cv2.THRESH_BINARY)
_, tpl_bin = cv2.threshold(template, 180, 255, cv2.THRESH_BINARY)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
roi_morph = cv2.morphologyEx(roi_bin, cv2.MORPH_OPEN, kernel)
tpl_morph = cv2.morphologyEx(tpl_bin, cv2.MORPH_OPEN, kernel)
th, tw = tpl_morph.shape[:2]
rh, rw = roi_morph.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
tpl_morph = cv2.resize(tpl_morph, (max(1, int(tw * scale)), max(1, int(th * scale))))
res = cv2.matchTemplate(roi_morph, tpl_morph, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results.append(("形态学开运算", "kernel(3,3)", float(max_val or 0.0)))
except Exception:
results.append(("形态学开运算", "kernel(3,3)", 0.0))
# 6. 形态学操作(闭运算)
try:
_, roi_bin = cv2.threshold(roi_img, 180, 255, cv2.THRESH_BINARY)
_, tpl_bin = cv2.threshold(template, 180, 255, cv2.THRESH_BINARY)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
roi_morph = cv2.morphologyEx(roi_bin, cv2.MORPH_CLOSE, kernel)
tpl_morph = cv2.morphologyEx(tpl_bin, cv2.MORPH_CLOSE, kernel)
th, tw = tpl_morph.shape[:2]
rh, rw = roi_morph.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
tpl_morph = cv2.resize(tpl_morph, (max(1, int(tw * scale)), max(1, int(th * scale))))
res = cv2.matchTemplate(roi_morph, tpl_morph, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results.append(("形态学闭运算", "kernel(3,3)", float(max_val or 0.0)))
except Exception:
results.append(("形态学闭运算", "kernel(3,3)", 0.0))
# 7. 原始图像(作为基准)
try:
th, tw = template.shape[:2]
rh, rw = roi_img.shape[:2]
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
template_scaled = cv2.resize(template, (max(1, int(tw * scale)), max(1, int(th * scale))))
else:
template_scaled = template
res = cv2.matchTemplate(roi_img, template_scaled, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results.append(("原始图像", "", float(max_val or 0.0)))
except Exception:
results.append(("原始图像", "", 0.0))
return results
def find_test_images():
"""查找所有直播间图标测试图片"""
pattern = os.path.join(DEBUG_CAPTURES_DIR, "**", "*直播间图标*.png")
images = glob.glob(pattern, recursive=True)
return images
def test_single_image(image_path, template):
"""测试单张图片的所有方法,返回结果字典"""
results = {}
try:
# 加载图片
test_img = load_image(image_path)
# 裁剪ROI
roi_img = crop_roi(test_img, ROI_X1, ROI_Y1, ROI_X2, ROI_Y2)
# 确保模板尺寸匹配
th, tw = template.shape[:2]
rh, rw = roi_img.shape[:2]
template_scaled = template.copy()
if th > rh or tw > rw:
scale = min(rh / max(th, 1), rw / max(tw, 1))
template_scaled = cv2.resize(template_scaled, (max(1, int(tw * scale)), max(1, int(th * scale))))
# 1. 测试固定阈值二值化
threshold_values = list(range(100, 250, 10))
for thresh in threshold_values:
try:
_, roi_bin = cv2.threshold(roi_img, thresh, 255, cv2.THRESH_BINARY)
_, tpl_bin = cv2.threshold(template_scaled, thresh, 255, cv2.THRESH_BINARY)
th_t, tw_t = tpl_bin.shape[:2]
rh_t, rw_t = roi_bin.shape[:2]
if th_t > rh_t or tw_t > rw_t:
scale = min(rh_t / max(th_t, 1), rw_t / max(tw_t, 1))
tpl_bin = cv2.resize(tpl_bin, (max(1, int(tw_t * scale)), max(1, int(th_t * scale))))
res = cv2.matchTemplate(roi_bin, tpl_bin, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
key = f"固定阈值二值化_{thresh}"
results[key] = float(max_val or 0.0)
except:
results[f"固定阈值二值化_{thresh}"] = 0.0
# 2. 测试自适应二值化
block_sizes = [3, 5, 7, 9, 11, 15]
c_values = [0, 1, 2, 3, 5]
for block_size in block_sizes:
for c in c_values:
try:
roi_bin = cv2.adaptiveThreshold(
roi_img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, block_size, c
)
tpl_bin = cv2.adaptiveThreshold(
template_scaled, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, block_size, c
)
th_t, tw_t = tpl_bin.shape[:2]
rh_t, rw_t = roi_bin.shape[:2]
if th_t > rh_t or tw_t > rw_t:
scale = min(rh_t / max(th_t, 1), rw_t / max(tw_t, 1))
tpl_bin = cv2.resize(tpl_bin, (max(1, int(tw_t * scale)), max(1, int(th_t * scale))))
res = cv2.matchTemplate(roi_bin, tpl_bin, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
key = f"自适应二值化_bs{block_size}_c{c}"
results[key] = float(max_val or 0.0)
except:
results[f"自适应二值化_bs{block_size}_c{c}"] = 0.0
# 3. 测试Canny边缘检测
low_thresholds = [20, 30, 40, 50, 60]
high_thresholds = [60, 80, 100, 120, 150, 180]
for low in low_thresholds:
for high in high_thresholds:
if high <= low:
continue
try:
roi_edges = cv2.Canny(roi_img, low, high)
tpl_edges = cv2.Canny(template_scaled, low, high)
th_t, tw_t = tpl_edges.shape[:2]
rh_t, rw_t = roi_edges.shape[:2]
if th_t > rh_t or tw_t > rw_t:
scale = min(rh_t / max(th_t, 1), rw_t / max(tw_t, 1))
tpl_edges = cv2.resize(tpl_edges, (max(1, int(tw_t * scale)), max(1, int(th_t * scale))))
res = cv2.matchTemplate(roi_edges, tpl_edges, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
key = f"Canny边缘检测_low{low}_high{high}"
results[key] = float(max_val or 0.0)
except:
results[f"Canny边缘检测_low{low}_high{high}"] = 0.0
# 4. 测试高级预处理方法
advanced_results = test_advanced_preprocessing(roi_img, template_scaled)
for method, params, score in advanced_results:
key = f"{method}_{params}" if params else method
results[key] = score
# 5. 测试原始图像
try:
res = cv2.matchTemplate(roi_img, template_scaled, cv2.TM_CCOEFF_NORMED)
_min_val, max_val, _min_loc, _max_loc = cv2.minMaxLoc(res)
results["原始图像"] = float(max_val or 0.0)
except:
results["原始图像"] = 0.0
except Exception as e:
print(f" 处理图片失败: {os.path.basename(image_path)} - {e}")
return {}
return results
def main():
print("=" * 80)
print("x图标模板匹配阈值优化测试(多图片批量测试)")
print("=" * 80)
print()
# 查找所有测试图片
print(f"扫描测试图片目录: {DEBUG_CAPTURES_DIR}")
test_images = find_test_images()
print(f"找到 {len(test_images)} 张直播间图标图片")
if len(test_images) == 0:
print("未找到测试图片!")
return
# 加载模板
print(f"\n加载模板: {TEMPLATE_PATH}")
try:
template = load_image(TEMPLATE_PATH)
print(f" 模板尺寸: {template.shape[1]}x{template.shape[0]}")
except Exception as e:
print(f" 错误: {e}")
return
print("\n" + "=" * 80)
print("开始批量测试...")
print("=" * 80)
# 统计所有方法在所有图片上的分数
method_scores = defaultdict(list) # {method_name: [score1, score2, ...]}
# 测试每张图片
for i, image_path in enumerate(test_images, 1):
print(f"\n[{i}/{len(test_images)}] 测试: {os.path.basename(image_path)}")
results = test_single_image(image_path, template)
# 收集分数
for method, score in results.items():
method_scores[method].append(score)
# 计算平均分数
print("\n" + "=" * 80)
print("【统计结果】所有方法的平均分数")
print("=" * 80)
avg_scores = []
for method, scores in method_scores.items():
avg_score = sum(scores) / len(scores) if scores else 0.0
max_score = max(scores) if scores else 0.0
min_score = min(scores) if scores else 0.0
avg_scores.append((method, avg_score, max_score, min_score, len(scores)))
# 按平均分数排序
avg_scores.sort(key=lambda x: x[1], reverse=True)
print(f"{'排名':<6} {'方法':<35} {'平均分':<10} {'最高分':<10} {'最低分':<10} {'图片数':<8} {'状态':<10}")
print("-" * 100)
for i, (method, avg_score, max_score, min_score, count) in enumerate(avg_scores[:30], 1): # 显示前30名
status = "✓ 通过" if avg_score >= 0.72 else "✗ 未通过"
print(f"{i:<6} {method:<35} {avg_score:<10.4f} {max_score:<10.4f} {min_score:<10.4f} {count:<8} {status:<10}")
# 推荐配置
print("\n" + "=" * 80)
print("【推荐配置】")
print("=" * 80)
if avg_scores:
best_method, best_avg, best_max, best_min, best_count = avg_scores[0]
print(f"最佳方法: {best_method}")
print(f"平均分数: {best_avg:.4f}")
print(f"最高分数: {best_max:.4f}")
print(f"最低分数: {best_min:.4f}")
print(f"测试图片数: {best_count}")
print(f"是否通过阈值(0.72): {'是' if best_avg >= 0.72 else '否'}")
# 解析最佳方法参数
if "固定阈值二值化" in best_method:
thresh = best_method.split("_")[-1]
print(f"\n推荐配置:")
print(f" 方法: 固定阈值二值化")
print(f" 阈值: {thresh}")
print(f" 模板匹配方法: TM_CCOEFF_NORMED")
elif "自适应二值化" in best_method:
parts = best_method.split("_")
bs = parts[1].replace("bs", "")
c = parts[2].replace("c", "")
print(f"\n推荐配置:")
print(f" 方法: 自适应二值化")
print(f" blockSize: {bs}")
print(f" C: {c}")
print(f" 模板匹配方法: TM_CCOEFF_NORMED")
elif "Canny边缘检测" in best_method:
parts = best_method.split("_")
low = parts[1].replace("low", "")
high = parts[2].replace("high", "")
print(f"\n推荐配置:")
print(f" 方法: Canny边缘检测")
print(f" low: {low}")
print(f" high: {high}")
print(f" 模板匹配方法: TM_CCOEFF_NORMED")
else:
print(f"\n推荐配置:")
print(f" 方法: {best_method}")
print(f" 模板匹配方法: TM_CCOEFF_NORMED")
print("\n测试完成!")
if __name__ == "__main__":
main()
"""
x图标模板匹配阈值优化测试脚本
测试不同阈值参数对模板匹配结果的影响
"""
import sys
import os
import cv2
import numpy as np
from pathlib import Path
import glob
from collections import defaultdict
# 设置UTF-8编码
if sys.platform == 'win32':
sys.stdout.reconfigure(encoding='utf-8')
# 测试图片路径(自动扫描debug_captures文件夹)
DEBUG_CAPTURES_DIR = r"pic/debug_captures"
TEMPLATE_PATH = r"assets/x.png"
# ROI坐标(基于1080x2400分辨率)
ROI_X1, ROI_Y1 = 972, 147
ROI_X2, ROI_Y2 = 1047, 216