File size: 8,443 Bytes
e4f8fe4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
import os
import cv2
import numpy as np
from ultralytics import YOLO
import tensorflow as tf
import tempfile

# ----------------- Helper Functions -----------------

def order_points(pts: np.ndarray) -> np.ndarray:
    """

    Orders 4 points in the order: top-left, top-right, bottom-right, bottom-left.

    """
    rect = np.zeros((4, 2), dtype="float32")
    s = pts.sum(axis=1)
    rect[0] = pts[np.argmin(s)]  # top-left
    rect[2] = pts[np.argmax(s)]  # bottom-right

    diff = np.diff(pts, axis=1)
    rect[1] = pts[np.argmin(diff)]  # top-right
    rect[3] = pts[np.argmax(diff)]  # bottom-left

    return rect


def crop_regions(image: np.ndarray, res, conf_threshold: float=0.6) -> list:
    """

    Crops and deskews regions based on OBB detection, returning (crop, x_min).

    Uses perspective transform to rotate the region upright.

    """
    regions = []
    if hasattr(res, 'obb') and res.obb is not None:
        polys = res.obb.xyxyxyxy.cpu().numpy()
        confs = res.obb.conf.cpu().numpy()
        for poly, conf in zip(polys, confs):
            if conf < conf_threshold:
                continue
            pts = poly.reshape(4, 2).astype(np.float32)
            # Order points and compute destination rectangle
            rect = order_points(pts)
            (tl, tr, br, bl) = rect
            widthA = np.linalg.norm(br - bl)
            widthB = np.linalg.norm(tr - tl)
            maxW = int(max(widthA, widthB))

            heightA = np.linalg.norm(tr - br)
            heightB = np.linalg.norm(tl - bl)
            maxH = int(max(heightA, heightB))

            dst = np.array([
                [0, 0],
                [maxW - 1, 0],
                [maxW - 1, maxH - 1],
                [0, maxH - 1]
            ], dtype="float32")

            # Perspective transform
            M = cv2.getPerspectiveTransform(rect, dst)
            warp = cv2.warpPerspective(image, M, (maxW, maxH))
            x_min = int(rect[:, 0].min())
            regions.append((warp, x_min))
    return regions

# ----------------- Main OCR Class -----------------
class TwoStageOCR:
    def __init__(

        self,

        box_model_path: str,

        yolo_model_path: str,

        cnn_model_path: str,

        image_size=(28, 28),

        conf_threshold=0.25

    ):
        # Stage 1: panel/region detector
        self.box_detector = YOLO(box_model_path)
        # Stage 2: digit detector for refined localization
        self.digit_detector = YOLO(yolo_model_path)
        # CNN for final classification
        self.cnn = tf.keras.models.load_model(cnn_model_path, compile=False)

        # Embedded class names for LeNet
        class_names = ['0','1','2','3','4','5','6','7','8','9','C','dot']
        self.inv_map = {i: label for i, label in enumerate(class_names)}

        self.image_size = image_size
        self.conf_threshold = conf_threshold

    def preprocess_crop(self, crop: np.ndarray) -> np.ndarray:
        gray = cv2.cvtColor(crop, cv2.COLOR_BGR2GRAY)
        resized = cv2.resize(gray, self.image_size)
        normed = resized.astype(np.float32) / 255.0
        return normed.reshape(1, *self.image_size, 1)

    def ocr_panel(self, panel: np.ndarray) -> str:
        """

        Detect digits in a cropped panel and classify using CNN.

        """
        res = self.digit_detector.predict(source=panel, verbose=False)[0]
        boxes = res.boxes.xyxy.cpu().numpy()
        confs = res.boxes.conf.cpu().numpy()
        # Filter by confidence
        mask = confs >= self.conf_threshold
        boxes = boxes[mask]
        if boxes.size == 0:
            return ""
        # Sort left-to-right
        boxes = boxes[np.argsort(boxes[:, 0])]

        digits = []
        for x1, y1, x2, y2 in boxes:
            c = panel[int(y1):int(y2), int(x1):int(x2)]
            inp = self.preprocess_crop(c)
            probs = self.cnn.predict(inp, verbose=False)
            idx = int(np.argmax(probs, axis=1)[0])
            label = self.inv_map[idx]
            digits.append(label)
        return ''.join(digits)

    def ocr_image(self, image_path: str) -> str:
        img = cv2.imread(image_path)
        if img is None:
            raise FileNotFoundError(f"Cannot read {image_path}")

        # Stage 1: detect and crop panels
        res_panels = self.box_detector.predict(source=img, verbose=False)[0]
        panels = crop_regions(img, res_panels)
        if not panels:
            return ""
        # Sort panels by x-coordinate
        panels = sorted(panels, key=lambda x: x[1])

        # Stage 2: OCR each panel
        results = []
        for panel_crop, _ in panels:
            text = self.ocr_panel(panel_crop)
            if text:
                results.append(text)
        return ' '.join(results)


# -------------------
# Example pipeline
# -------------------

if __name__ == '__main__':
    box_model = 'Models/res_temp_box_v3.pt'
    temp_yolo_model = 'Models/temp_detect_v3.pt'
    temp_cnn_model  = 'Models/lenet7seg.h5'
    res_yolo_model = 'Models/res_detect_v4.pt'
    res_cnn_model  = 'Models/lenet_res_v4.h5'

    # Initialize both OCRs
    temp_ocr = TwoStageOCR(
        box_model_path=box_model,
        yolo_model_path=temp_yolo_model,
        cnn_model_path=temp_cnn_model,
        image_size=(28,28),
        conf_threshold=0.3
    )
    from Lenet_res import YoloLeNetOCR 
    res_ocr = YoloLeNetOCR(
        yolo_model_path=res_yolo_model,
        lenet_model_path=res_cnn_model, 
        image_size=(28,28),
        conf_threshold=0.5
    )

    input_dir = 'cr_test'
    for fname in os.listdir(input_dir):
        if not fname.lower().endswith(('.png','.jpg','.jpeg')):
            continue
        full = os.path.join(input_dir, fname)
        img = cv2.imread(full)
        if img is None:
            print(f"Cannot read {full}")
            continue

        # Detect panels (res or temp)
        box_detector = temp_ocr.box_detector
        res_panels = box_detector.predict(source=img, verbose=False)[0]
        if not hasattr(res_panels, 'obb') or res_panels.obb is None:
            print(f"{fname} -> No panels detected")
            continue

        polys = res_panels.obb.xyxyxyxy.cpu().numpy()   
        confs = res_panels.obb.conf.cpu().numpy()
        class_ids = res_panels.obb.cls.cpu().numpy().astype(int)
        class_names = box_detector.model.names  # Should be ['res', 'temp']

        for poly, conf, cls_id in zip(polys, confs, class_ids):
            if conf < 0.3:
                continue
            pts = poly.reshape(4, 2).astype(np.float32)
            rect = order_points(pts)  # <-- FIXED HERE
            (tl, tr, br, bl) = rect
            widthA = np.linalg.norm(br - bl)
            widthB = np.linalg.norm(tr - tl)
            maxW = int(max(widthA, widthB))
            heightA = np.linalg.norm(tr - br)
            heightB = np.linalg.norm(tl - bl)
            maxH = int(max(heightA, heightB))
            dst = np.array([
                [0, 0],
                [maxW - 1, 0],
                [maxW - 1, maxH - 1],
                [0, maxH - 1]
            ], dtype="float32")
            M = cv2.getPerspectiveTransform(rect, dst)
            crop = cv2.warpPerspective(img, M, (maxW, maxH))

            class_name = class_names[cls_id]
            if class_name == 'temp':
                raw = temp_ocr.ocr_panel(crop)
                # Remove 'C' if present
                temp_digits = raw.replace('C', '')
                # Only format if at least 2 digits
                if len(temp_digits) > 1:
                    formatted = temp_digits[:-1] + '.' + temp_digits[-1] + '°C'
                else:
                    formatted = temp_digits + '°C'
                print(f"{fname} [TEMP] -> {formatted}")
            elif class_name == 'res':
                # Save crop to a temporary file
                with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as tmp:
                    cv2.imwrite(tmp.name, crop)
                    tmp_path = tmp.name
                raw = res_ocr.ocr_image(tmp_path)
                os.remove(tmp_path)  # Clean up temp file
                raw = raw.replace("dot", ".")  
                print(f"{fname} [RES] -> {raw}")