import cv2
import os
import difflib
import numpy as np
from paddleocr import PaddleOCR

def similarity(s1, s2):
    return difflib.SequenceMatcher(None, s1, s2).ratio()

def optimize_unwarp():
    target = "威凯检测技术有限公司"
    img_path = "seal_cropped.png"
    img = cv2.imread(img_path)
    if img is None: return
    
    h_orig, w_orig = img.shape[:2]
    cx_base, cy_base = w_orig // 2, h_orig // 2
    radius = min(cx_base, cy_base)
    
    ocr = PaddleOCR(use_angle_cls=True, lang='ch')
    
    # Preprocess: remove red?
    # For now just use original to be safe.
    
    # Test a small grid around the center
    best_overall_sim = 0.0
    best_params = {}
    
    for dx in [-10, -5, 0, 5, 10]:
        for dy in [-10, -5, 0, 5, 10]:
            cx, cy = cx_base + dx, cy_base + dy
            # We use a larger factor to be sure
            out_theta = 1500
            out_r = radius
            
            unwarped = cv2.warpPolar(img, (out_r, out_theta), (cx, cy), radius, cv2.WARP_POLAR_LINEAR)
            final = cv2.transpose(unwarped)
            
            # Try a few different annuli
            for r_start, r_end in [(0.7, 0.9), (0.75, 0.95), (0.8, 1.0)]:
                strip = final[int(out_r * r_start):int(out_r * r_end), :]
                
                # Padding is essential for small strips
                padded = cv2.copyMakeBorder(strip, 40, 40, 20, 20, cv2.BORDER_CONSTANT, value=[255, 255, 255])
                
                res = ocr.ocr(padded)
                
                current_best_text = ""
                current_best_sim = 0.0
                if res:
                    for page in res:
                        if 'rec_texts' in page:
                            for t in page['rec_texts']:
                                ct = t.replace(" ", "")
                                s = similarity(target, ct)
                                if s > current_best_sim:
                                    current_best_sim = s
                                    current_best_text = ct
                
                if current_best_sim > best_overall_sim:
                    best_overall_sim = current_best_sim
                    best_params = {
                        "dx": dx, "dy": dy, "r_start": r_start, "r_end": r_end,
                        "text": current_best_text, "sim": current_best_sim
                    }
                    print(f"NEW BEST: {best_overall_sim:.4f} | {best_params}")

    print(f"\nFINAL BEST: {best_overall_sim:.4f}")
    print(best_params)

if __name__ == "__main__":
    optimize_unwarp()