File size: 4,481 Bytes
77da9e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
"""
OCR Handler - OCR-only Processing

This module provides OCR-only functionality that bypasses the full detection pipeline.
Useful for cases where you only need text extraction without RF-DETR/CLIP analysis.
"""

import torch
import cv2
import numpy as np
from PIL import Image
from typing import Union, List, Dict, Tuple
from pathlib import Path
import easyocr

from detection.image_utils import load_image


def process_ocr_only(
    image: Union[str, Path, np.ndarray, Image.Image],
    gpu: bool = None
) -> List[Dict]:
    """
    Run OCR across the full image and return detections
    
    This bypasses RF-DETR/CLIP and runs EasyOCR directly on the image.
    
    Args:
        image: Input image (path, PIL Image, or numpy array)
        gpu: Whether to use GPU. If None, auto-detects CUDA availability.
        
    Returns:
        List of detections with keys:
            - box: Dict with x1, y1, x2, y2 coordinates
            - confidence: OCR confidence score (float)
            - class_id: None (no classification)
            - class_name: "" (no classification)
            - text: Extracted text string
            - description: "" (no description)
    """
    # Load image
    img_array = load_image(image)
    
    # Initialize OCR reader
    if gpu is None:
        gpu = torch.cuda.is_available()
    reader = easyocr.Reader(['en', 'fr'], gpu=gpu)
    
    # Run OCR - detail=1 returns [ [ (x,y)...4 points ], text, conf ]
    ocr_results = reader.readtext(img_array, detail=1)
    
    # Convert to standard detection format
    detections = []
    for entry in ocr_results:
        if not isinstance(entry, (list, tuple)) or len(entry) < 3:
            continue
        quad, text, conf = entry[0], entry[1], entry[2]
        if not isinstance(text, str) or not text.strip():
            continue
        
        # Convert quadrilateral to bounding box
        xs = [p[0] for p in quad]
        ys = [p[1] for p in quad]
        box = {
            "x1": float(int(min(xs))),
            "y1": float(int(min(ys))),
            "x2": float(int(max(xs))),
            "y2": float(int(max(ys)))
        }
        
        detections.append({
            "box": box,
            "confidence": float(conf) if conf is not None else 1.0,
            "class_id": None,
            "class_name": "",
            "text": text.strip(),
            "description": ""
        })
    
    return detections


def annotate_ocr_detections(
    image: Union[str, Path, np.ndarray, Image.Image],
    detections: List[Dict],
    thickness: int = 2,
    return_format: str = "pil"
) -> Union[Image.Image, np.ndarray]:
    """
    Annotate image with OCR detection boxes and text labels
    
    Args:
        image: Input image (path, PIL Image, or numpy array)
        detections: List of detections from process_ocr_only()
        thickness: Line thickness for bounding boxes
        return_format: "pil" for PIL Image or "numpy" for numpy array
        
    Returns:
        Annotated image as PIL Image or numpy array
    """
    # Load image
    img_array = load_image(image)
    
    # Convert to BGR for OpenCV
    img_bgr = cv2.cvtColor(img_array, cv2.COLOR_RGB2BGR)
    
    # Draw each detection
    for det in detections:
        x1 = int(det["box"]["x1"])
        y1 = int(det["box"]["y1"])
        x2 = int(det["box"]["x2"])
        y2 = int(det["box"]["y2"])
        
        # Draw bounding box
        cv2.rectangle(img_bgr, (x1, y1), (x2, y2), (0, 255, 0), thickness)
        
        # Draw text label
        text = det.get("text", "")
        if text:
            (tw, th), bl = cv2.getTextSize(text, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1)
            ty = max(y1 - 10, th + 10)
            
            # Draw text background
            cv2.rectangle(
                img_bgr,
                (x1, ty - th - bl - 4),
                (x1 + tw + 6, ty + bl - 4),
                (0, 180, 0),  # Darker green
                -1
            )
            
            # Draw text
            cv2.putText(
                img_bgr,
                text,
                (x1 + 3, ty - bl - 2),
                cv2.FONT_HERSHEY_SIMPLEX,
                0.5,
                (255, 255, 255),
                1
            )
    
    # Convert back to RGB
    img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
    
    # Return in requested format
    if return_format.lower() == "pil":
        return Image.fromarray(img_rgb)
    else:
        return img_rgb