forked from pradana.aumars/videocr
Merge pull request 'Add additional image processing' (#1) from Yun/videocr:master into master
Reviewed-on: pradana.aumars/videocr#1
This commit is contained in:
commit
b005e36fcd
@ -18,6 +18,8 @@ class Video:
|
||||
num_frames: int
|
||||
fps: float
|
||||
height: int
|
||||
width: int
|
||||
resize_dim: List[int]
|
||||
pred_frames: List[PredictedFrame]
|
||||
pred_subs: List[PredictedSubtitle]
|
||||
|
||||
@ -27,6 +29,9 @@ class Video:
|
||||
self.num_frames = int(v.get(cv2.CAP_PROP_FRAME_COUNT))
|
||||
self.fps = v.get(cv2.CAP_PROP_FPS)
|
||||
self.height = int(v.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
||||
self.width = int(v.get(cv2.CAP_PROP_FRAME_WIDTH))
|
||||
scale_percent = 47 # apparently 32 pixels is the optimal character height for tesseract.
|
||||
self.resize_dim=(int(self.width * scale_percent/100), int(self.height * scale_percent/100))
|
||||
|
||||
def run_ocr(self, lang: str, time_start: str, time_end: str,
|
||||
conf_threshold: int, use_fullframe: bool) -> None:
|
||||
@ -56,8 +61,16 @@ class Video:
|
||||
if not self.use_fullframe:
|
||||
# only use bottom half of the frame by default
|
||||
img = img[self.height // 2:, :]
|
||||
img = cv2.bitwise_not(cv2.bitwise_and(img, img, mask=cv2.inRange(img, (190, 190, 190), (255, 255, 255))))
|
||||
config = '--tessdata-dir "{}"'.format(constants.TESSDATA_DIR)
|
||||
# dilate and resize
|
||||
img=cv2.resize(cv2.dilate(img, np.ones((2, 2), np.uint8)), self.resize_dim, interpolation=cv2.INTER_AREA)
|
||||
|
||||
# mask to filter out non gray-like pixels/pixels that are not bright enough
|
||||
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
|
||||
color_mask = cv2.inRange(hsv, (0, 0, 190), (179, 20, 255))
|
||||
|
||||
# apply mask, inverse image so it's black text on white background, add borders to top and bottom
|
||||
img = cv2.copyMakeBorder(cv2.bitwise_not(cv2.bitwise_and(img, img, mask=color_mask)), 10, 10, 0, 0, cv2.BORDER_CONSTANT, None, (255,255,255)
|
||||
config = '--tessdata-dir "{}" --psm 7 -c preserve_interword_spaces=1'.format(constants.TESSDATA_DIR)
|
||||
try:
|
||||
return pytesseract.image_to_data(img, lang=self.lang, config=config)
|
||||
except Exception as e:
|
||||
|
Loading…
Reference in New Issue
Block a user