claudetools/review_best_plates.py

"""
Identify the best license plate candidates from extraction results
Filter by ideal aspect ratio (2-5) and larger area
"""

import re
from pathlib import Path

def parse_summary(summary_path):
    """Parse summary.txt to find best candidates"""
    candidates = []

    with open(summary_path, 'r') as f:
        content = f.read()

    # Parse each candidate line
    pattern = r'Time: ([\d.]+)s \| Candidate #(\d+) \| Aspect Ratio: ([\d.]+) \| Area: (\d+)'

    for match in re.finditer(pattern, content):
        timestamp = float(match.group(1))
        candidate_num = int(match.group(2))
        aspect_ratio = float(match.group(3))
        area = int(match.group(4))

        # Score candidates based on ideal license plate characteristics
        # Ideal aspect ratio: 3-4.5 (most US license plates)
        # Prefer larger areas (closer to camera)
        ar_score = 0
        if 2.5 <= aspect_ratio <= 5.0:
            # Best score for aspect ratio between 3-4.5
            if 3.0 <= aspect_ratio <= 4.5:
                ar_score = 100
            else:
                ar_score = 50

        # Area score (normalize to 0-100)
        area_score = min(area / 500, 100)  # Scale area

        # Combined score
        total_score = (ar_score * 0.6) + (area_score * 0.4)

        candidates.append({
            'timestamp': timestamp,
            'candidate': candidate_num,
            'aspect_ratio': aspect_ratio,
            'area': area,
            'score': total_score
        })

    return candidates

def main():
    summary_path = Path("D:/Scratchpad/pickup_truck_25-30s/summary.txt")
    output_dir = Path("D:/Scratchpad/pickup_truck_25-30s")

    print("[INFO] Analyzing license plate candidates...")
    candidates = parse_summary(summary_path)

    # Sort by score
    candidates.sort(key=lambda x: x['score'], reverse=True)

    # Show top 20 candidates
    print("\n" + "=" * 80)
    print("TOP 20 LICENSE PLATE CANDIDATES")
    print("=" * 80)
    print(f"{'Rank':<6} {'Time':<10} {'Cand':<6} {'AR':<8} {'Area':<10} {'Score':<8} {'Files'}")
    print("-" * 80)

    for idx, candidate in enumerate(candidates[:20], 1):
        timestamp = candidate['timestamp']
        cand_num = candidate['candidate']
        ar = candidate['aspect_ratio']
        area = candidate['area']
        score = candidate['score']

        # Check which files exist for this candidate
        frame_name = f"frame_{timestamp:.2f}s"
        base_pattern = f"{frame_name}_plate_{cand_num}_"

        # Count enhancement files
        enhancement_files = list(output_dir.glob(f"{base_pattern}*.jpg"))
        enhancement_count = len([f for f in enhancement_files if '_raw' not in f.name])

        print(f"{idx:<6} {timestamp:<10.2f} {cand_num:<6} {ar:<8.2f} {area:<10} {score:<8.1f} {enhancement_count} enhanced")

    # Create recommendation file
    recommendation_path = output_dir / "RECOMMENDATIONS.txt"
    with open(recommendation_path, 'w') as f:
        f.write("LICENSE PLATE EXTRACTION - TOP CANDIDATES\n")
        f.write("=" * 80 + "\n\n")
        f.write("These are the top 20 most likely license plate candidates based on:\n")
        f.write("- Aspect ratio (ideal: 3.0-4.5 for US plates)\n")
        f.write("- Area size (larger = closer to camera)\n\n")
        f.write("REVIEW THESE FILES FIRST:\n")
        f.write("-" * 80 + "\n\n")

        for idx, candidate in enumerate(candidates[:20], 1):
            timestamp = candidate['timestamp']
            cand_num = candidate['candidate']
            ar = candidate['aspect_ratio']
            area = candidate['area']
            score = candidate['score']

            f.write(f"RANK {idx}: Time {timestamp:.2f}s - Candidate #{cand_num}\n")
            f.write(f"  Aspect Ratio: {ar:.2f} | Area: {area} | Score: {score:.1f}\n")
            f.write(f"  Files to review:\n")

            frame_name = f"frame_{timestamp:.2f}s"

            # List specific enhancement files to check
            enhancements = [
                f"{frame_name}_detection_{cand_num}.jpg  (shows detection box on frame)",
                f"{frame_name}_plate_{cand_num}_high_contrast.jpg  (best for dark plates)",
                f"{frame_name}_plate_{cand_num}_extreme_sharp.jpg  (best for clarity)",
                f"{frame_name}_plate_{cand_num}_adaptive_thresh.jpg  (best for OCR)",
                f"{frame_name}_plate_{cand_num}_bilateral_sharp.jpg  (balanced enhancement)",
            ]

            for enhancement in enhancements:
                f.write(f"    - {enhancement}\n")

            f.write("\n")

        f.write("\n" + "=" * 80 + "\n")
        f.write("ENHANCEMENT TYPES EXPLAINED:\n")
        f.write("-" * 80 + "\n")
        f.write("- detection_X.jpg: Shows where the plate was detected on the frame\n")
        f.write("- high_contrast.jpg: Best for dark/low-contrast plates\n")
        f.write("- extreme_sharp.jpg: Best for overall clarity and readability\n")
        f.write("- adaptive_thresh.jpg: Black/white threshold - best for OCR\n")
        f.write("- bilateral_sharp.jpg: Noise reduction + sharpening\n")
        f.write("- unsharp_mask.jpg: Professional-grade sharpening\n")
        f.write("- bright_contrast.jpg: Brightness + contrast boost\n")

    print("\n[SUCCESS] Analysis complete!")
    print(f"[INFO] Recommendations saved to: {recommendation_path}")
    print("\n[NEXT STEPS]")
    print("1. Open the output directory in File Explorer:")
    print(f"   {output_dir}")
    print("2. Read RECOMMENDATIONS.txt for the best candidates")
    print("3. Start with Rank 1, review the enhancement files listed")
    print("4. The 'extreme_sharp' and 'adaptive_thresh' versions usually work best")

if __name__ == "__main__":
    main()