Files
rocr/rusto-rs/create_test_images.py
ssw c303c315e5 rocr: rusto-rs 源码入库(上游 v0.2.5 基线 57609fc + 本地补丁)
- 源码 316 文件收编为 rusto-rs/ 文件夹,随本仓库管理
- 移除内部 .git(补丁已存档 docs/rusto-rs-metal-patch-20260912.patch,上游可随时 re-clone)
- target/ 与 docs/.yarn 缓存不入库;本地改动:ROCR_* 环境变量调优 + Metal 链接修复 + 预处理测试脚本
2026-09-12 17:30:04 +08:00

178 lines
6.5 KiB
Python

#!/usr/bin/env python3
"""
Create test images for orientation and rectification testing.
Uses example1.png from models/test_images as base image.
Outputs to models/test_images directory.
"""
from PIL import Image, ImageFilter
import numpy as np
import sys
from pathlib import Path
# Configuration
BASE_IMAGE = "models/test_images/example1.png"
OUTPUT_DIR = "models/test_images"
def create_rotated_image(input_path, output_path, angle):
"""Create a high-quality rotated version of the image."""
img = Image.open(input_path)
# Use BICUBIC for better quality
rotated = img.rotate(angle, expand=True, fillcolor='white', resample=Image.BICUBIC)
rotated.save(output_path, quality=95)
print(f"✓ Created {output_path.name} (rotated {angle}°)")
return rotated
def create_curved_image(input_path, output_path):
"""Create a curved/distorted version for rectification testing."""
img = Image.open(input_path)
img_array = np.array(img)
height, width = img_array.shape[:2]
# Create a mesh grid
x = np.arange(width)
y = np.arange(height)
xx, yy = np.meshgrid(x, y)
# Apply wave distortion
amplitude = 20 # pixels
frequency = 2 * np.pi / width
# Horizontal wave
xx_distorted = xx + amplitude * np.sin(frequency * yy)
# Clip to valid range
xx_distorted = np.clip(xx_distorted, 0, width - 1).astype(int)
# Create distorted image
distorted = np.zeros_like(img_array)
for i in range(height):
for j in range(width):
distorted[i, j] = img_array[i, xx_distorted[i, j]]
result = Image.fromarray(distorted)
result.save(output_path, quality=95)
print(f"✓ Created {output_path.name} (curved distortion)")
return result
def create_perspective_distorted(input_path, output_path):
"""Create perspective-distorted image for rectification testing."""
img = Image.open(input_path)
width, height = img.size
# Define perspective transform coefficients
# This creates a trapezoidal effect
coeffs = find_coeffs(
[(0, 0), (width, 0), (width, height), (0, height)], # original corners
[(50, 30), (width-30, 50), (width-50, height-30), (30, height-50)] # distorted corners
)
distorted = img.transform(
(width, height),
Image.PERSPECTIVE,
coeffs,
Image.BICUBIC
)
distorted.save(output_path, quality=95)
print(f"✓ Created {output_path.name} (perspective distortion)")
return distorted
def find_coeffs(source_coords, target_coords):
"""Find perspective transform coefficients."""
matrix = []
for s, t in zip(source_coords, target_coords):
matrix.append([t[0], t[1], 1, 0, 0, 0, -s[0]*t[0], -s[0]*t[1]])
matrix.append([0, 0, 0, t[0], t[1], 1, -s[1]*t[0], -s[1]*t[1]])
A = np.array(matrix, dtype=float)
B = np.array(source_coords).reshape(8)
try:
res = np.linalg.solve(A, B)
return np.array(res).tolist()
except:
# Fallback to identity-like transform
return [1, 0, 0, 0, 1, 0, 0, 0]
def create_noisy_rotated(input_path, output_path, angle):
"""Create rotated image with minimal noise to preserve readability."""
img = Image.open(input_path)
# Rotate with high quality settings
rotated = img.rotate(angle, expand=True, fillcolor='white', resample=Image.BICUBIC)
# Add very slight noise (reduced from 10 to 3 for better quality)
img_array = np.array(rotated)
noise = np.random.normal(0, 3, img_array.shape).astype(np.int16)
noisy = np.clip(img_array.astype(np.int16) + noise, 0, 255).astype(np.uint8)
result = Image.fromarray(noisy)
result.save(output_path, quality=95)
print(f"✓ Created {output_path.name} (rotated {angle}° with minimal noise)")
return result
def main():
# Check if input file exists
input_file = Path(BASE_IMAGE)
if not input_file.exists():
print(f"Error: Input file not found: {input_file}")
print("Please ensure models/test_images/example1.png exists")
sys.exit(1)
# Check if output directory exists
output_dir = Path(OUTPUT_DIR)
if not output_dir.exists():
print(f"Error: Output directory not found: {output_dir}")
print("Please ensure models/test_images/ directory exists")
sys.exit(1)
print("=== Creating Test Images ===\n")
print(f"Base image: {input_file}")
print(f"Output directory: {output_dir}\n")
# 1. Orientation test images (0°, 90°, 180°, 270°)
print("1. Orientation Classification Tests:")
create_rotated_image(input_file, output_dir / "example1_rotate_90.png", 90)
create_rotated_image(input_file, output_dir / "example1_rotate_180.png", 180)
create_rotated_image(input_file, output_dir / "example1_rotate_270.png", 270)
# 2. Slight rotation (5-15 degrees) for angle correction
# Using minimal noise to preserve readability
print("\n2. Small Rotation Tests (with minimal noise):")
create_noisy_rotated(input_file, output_dir / "example1_rotate_5.png", 5)
create_noisy_rotated(input_file, output_dir / "example1_rotate_15.png", 15)
create_noisy_rotated(input_file, output_dir / "example1_rotate_minus10.png", -10)
# 3. Rectification test images (no barrel - removed as requested)
print("\n3. Rectification Tests:")
create_curved_image(input_file, output_dir / "example1_curved.png")
create_perspective_distorted(input_file, output_dir / "example1_perspective.png")
# 4. Combined tests
print("\n4. Combined Tests:")
# Create curved then rotate it
temp_curved_path = output_dir / "temp_curved.png"
temp_curved = create_curved_image(input_file, temp_curved_path)
temp_curved_rotated = temp_curved.rotate(90, expand=True, fillcolor='white', resample=Image.BICUBIC)
temp_curved_rotated.save(output_dir / "example1_curved_rotated.png", quality=95)
print(f"✓ Created example1_curved_rotated.png (curved + rotated)")
# Clean up temp file
if temp_curved_path.exists():
temp_curved_path.unlink()
print(f"\n=== Done! ===")
# Count new test images (exclude original example1.png)
test_images = [f for f in output_dir.glob('example1_*.png')]
print(f"Created {len(test_images)} test images in {output_dir}/")
print("\nTest images:")
for img in sorted(test_images):
print(f" - {img.name}")
print("\nUsage:")
print(f" cargo run --release --example ocr_with_fixes")
print(f" cargo run --release --example ocr_without_fixes")
if __name__ == "__main__":
main()