text_detect/core/text_detect.py
2026-03-08 12:19:45 +03:00

256 lines
9.2 KiB
Python

# -*- coding: utf-8 -*-
"""Text region detection using OpenCV EAST with graceful fallback."""
import os
import sys
import numpy as np
try:
import cv2
except ImportError: # pragma: no cover
cv2 = None
class TextDetector(object):
def __init__(self, east_model_path, score_threshold=0.4, use_cuda=False):
self.east_model_path = east_model_path
self.score_threshold = score_threshold
self.use_cuda = use_cuda
self.net = None
self.east_available = False
self._load()
def _load(self):
if cv2 is None:
print('[WARN] OpenCV not installed; EAST disabled', file=sys.stderr)
return
if not os.path.exists(self.east_model_path):
print('[INFO] EAST model not found; using OpenCV contour fallback', file=sys.stderr)
return
try:
self.net = cv2.dnn.readNet(self.east_model_path)
if self.use_cuda and hasattr(cv2, "cuda") and cv2.cuda.getCudaEnabledDeviceCount() > 0:
try:
self.net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
self.net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)
print('[INFO] EAST using CUDA backend/target', file=sys.stderr)
except Exception as exc:
print('[WARN] Failed to enable CUDA for EAST: {}'.format(exc), file=sys.stderr)
self.east_available = True
print('[INFO] EAST model loaded: {}'.format(self.east_model_path))
except Exception as exc:
print('[WARN] Failed to load EAST model: {}'.format(exc), file=sys.stderr)
self.east_available = False
def detect(self, img_rgb):
h, w = img_rgb.shape[:2]
if self.east_available:
boxes = self._detect_with_east(img_rgb)
if boxes:
return boxes, True
boxes = self._detect_with_cv_fallback(img_rgb)
if boxes:
return boxes, False
return [(_int0(0), _int0(0), _int0(w), _int0(h))], False
def _detect_with_east(self, img_rgb):
h, w = img_rgb.shape[:2]
# EAST expects width/height divisible by 32
new_w = 320
new_h = 320
blob = cv2.dnn.blobFromImage(img_rgb, 1.0, (new_w, new_h),
(123.68, 116.78, 103.94), swapRB=True, crop=False)
self.net.setInput(blob)
scores, geometry = self.net.forward([
"feature_fusion/Conv_7/Sigmoid",
"feature_fusion/concat_3"
])
rectangles, confidences = self._decode(scores, geometry, self.score_threshold)
indices = cv2.dnn.NMSBoxes(rectangles, confidences, self.score_threshold, 0.4)
boxes = []
rW = float(w) / float(new_w)
rH = float(h) / float(new_h)
if len(indices) > 0:
for i in indices.flatten():
x, y, bw, bh = rectangles[i]
x1 = int(x * rW)
y1 = int(y * rH)
x2 = int((x + bw) * rW)
y2 = int((y + bh) * rH)
boxes.append((_int0(x1), _int0(y1), _int0(x2), _int0(y2)))
return self._normalize_boxes(boxes, w, h, max_boxes=16)
def _decode(self, scores, geometry, score_thresh):
num_rows, num_cols = scores.shape[2:4]
rectangles = []
confidences = []
for y in range(num_rows):
scores_data = scores[0, 0, y]
x0 = geometry[0, 0, y]
x1 = geometry[0, 1, y]
x2 = geometry[0, 2, y]
x3 = geometry[0, 3, y]
angles = geometry[0, 4, y]
for x in range(num_cols):
score = scores_data[x]
if score < score_thresh:
continue
offset_x = x * 4.0
offset_y = y * 4.0
angle = angles[x]
cos = np.cos(angle)
sin = np.sin(angle)
h = x0[x] + x2[x]
w = x1[x] + x3[x]
end_x = int(offset_x + (cos * x1[x]) + (sin * x2[x]))
end_y = int(offset_y - (sin * x1[x]) + (cos * x2[x]))
start_x = int(end_x - w)
start_y = int(end_y - h)
rectangles.append((start_x, start_y, int(w), int(h)))
confidences.append(float(score))
return rectangles, confidences
def _detect_with_cv_fallback(self, img_rgb):
if cv2 is None:
return []
try:
gray = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2GRAY)
except Exception:
return []
h, w = gray.shape[:2]
img_area = float(max(1, h * w))
min_area = max(160.0, img_area * 0.00035)
boxes = []
# Pass #1: gradient + morphology for horizontal text lines.
try:
grad_x = cv2.Sobel(gray, ddepth=cv2.CV_32F, dx=1, dy=0, ksize=3)
grad_x = cv2.convertScaleAbs(grad_x)
grad_x = cv2.GaussianBlur(grad_x, (3, 3), 0)
_, bin1 = cv2.threshold(grad_x, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
k_close = cv2.getStructuringElement(cv2.MORPH_RECT, (25, 5))
k_dilate = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 3))
bin1 = cv2.morphologyEx(bin1, cv2.MORPH_CLOSE, k_close, iterations=1)
bin1 = cv2.dilate(bin1, k_dilate, iterations=1)
boxes.extend(self._boxes_from_mask(bin1, w, h, min_area=min_area))
except Exception:
pass
# Pass #2: adaptive threshold for low-contrast / noisy scenes.
try:
bin2 = cv2.adaptiveThreshold(
gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 31, 15
)
k_close = cv2.getStructuringElement(cv2.MORPH_RECT, (15, 3))
k_dilate = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
bin2 = cv2.morphologyEx(bin2, cv2.MORPH_CLOSE, k_close, iterations=1)
bin2 = cv2.dilate(bin2, k_dilate, iterations=1)
boxes.extend(self._boxes_from_mask(bin2, w, h, min_area=min_area))
except Exception:
pass
return self._normalize_boxes(boxes, w, h, max_boxes=18)
def _boxes_from_mask(self, mask, image_w, image_h, min_area):
out = []
contours_result = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
contours = contours_result[0] if len(contours_result) == 2 else contours_result[1]
min_h = max(10, int(image_h * 0.015))
min_w = max(16, int(image_w * 0.02))
for cnt in contours:
x, y, bw, bh = cv2.boundingRect(cnt)
area = float(bw * bh)
if area < min_area:
continue
if bw < min_w or bh < min_h:
continue
aspect = float(bw) / float(max(1, bh))
if aspect < 0.7 or aspect > 45.0:
continue
contour_area = float(cv2.contourArea(cnt))
fill_ratio = contour_area / float(max(1, bw * bh))
if fill_ratio < 0.03:
continue
pad_x = max(2, int(bw * 0.08))
pad_y = max(2, int(bh * 0.30))
x1 = max(0, x - pad_x)
y1 = max(0, y - pad_y)
x2 = min(image_w, x + bw + pad_x)
y2 = min(image_h, y + bh + pad_y)
out.append((_int0(x1), _int0(y1), _int0(x2), _int0(y2)))
return out
def _normalize_boxes(self, boxes, image_w, image_h, max_boxes=16):
norm = []
for box in boxes or []:
x1, y1, x2, y2 = _clip_box(box, image_w, image_h)
if x2 <= x1 or y2 <= y1:
continue
norm.append((x1, y1, x2, y2))
if not norm:
return []
kept = _nms_by_iou(norm, iou_thresh=0.45)
kept = sorted(kept, key=lambda b: ((b[3] - b[1]) * (b[2] - b[0])), reverse=True)
if max_boxes and len(kept) > max_boxes:
kept = kept[:max_boxes]
kept = sorted(kept, key=lambda b: (b[1], b[0]))
return kept
def _int0(val):
try:
return int(val)
except Exception:
return 0
def _clip_box(box, image_w, image_h):
x1, y1, x2, y2 = box
x1 = max(0, min(_int0(x1), _int0(image_w)))
y1 = max(0, min(_int0(y1), _int0(image_h)))
x2 = max(0, min(_int0(x2), _int0(image_w)))
y2 = max(0, min(_int0(y2), _int0(image_h)))
return x1, y1, x2, y2
def _nms_by_iou(boxes, iou_thresh=0.45):
if not boxes:
return []
ranked = sorted(boxes, key=_box_area, reverse=True)
keep = []
for candidate in ranked:
should_keep = True
for chosen in keep:
if _iou(candidate, chosen) >= iou_thresh:
should_keep = False
break
if should_keep:
keep.append(candidate)
return keep
def _box_area(box):
x1, y1, x2, y2 = box
return max(0, x2 - x1) * max(0, y2 - y1)
def _iou(a, b):
ax1, ay1, ax2, ay2 = a
bx1, by1, bx2, by2 = b
xx1 = max(ax1, bx1)
yy1 = max(ay1, by1)
xx2 = min(ax2, bx2)
yy2 = min(ay2, by2)
iw = max(0, xx2 - xx1)
ih = max(0, yy2 - yy1)
inter = iw * ih
if inter <= 0:
return 0.0
union = float(_box_area(a) + _box_area(b) - inter)
if union <= 0:
return 0.0
return inter / union