Recherche avancée

Sur d’autres sites (9142)

• ffmpeg - make pixels transparent depending on how black they are

  8 juillet 2022, par TheForgot3n1

  Fully black pixels should be fully transparent. Grey pixels should be semi-transparent. White pixels should not be transparent at all. Is this possible to achieve in ffmpeg ?

  


• OpenCV Encoding to H264 changing original RGB pixel values for gray images

  13 janvier 2020, par CristoJV

  I have the following issue :
  I’m creating a uniform gray color video (for testing) using OpenCV VideoWriter. The output video will reproduce a constant image where all the pixels must have the same value x (25, 51, 76,... and so on).
  When I generate the video using MJPG Encoder :

  vw = cv2.VideoWriter('./videos/input/gray1.mp4',
  
          cv2.VideoWriter_fourcc(*'MJPG'),
  
          fps,(resolution[1],resolution[0]))

  and read the output using the VideoCapture class, everything just works fine. I got a frame array with all pixel values set to (25,51,76 and so on).
  However when I generate the video using HEV1 (H.265) or also H264 :

  vw = cv2.VideoWriter('./videos/input/gray1.mp4',
  
          cv2.VideoWriter_fourcc(*'HEV1'),
  
          fps,(resolution[1],resolution[0]))

  I run into the following issue. The frame I got in BGR format follows the next configuration :

  • The blue channel value is the expected value (x) minus 4 (25-4=21, 51-4=47, 76-4=72, and so on).
  • The green channel is the expected value (x) minus 1 (25-1=24, 51-1=50, 76-1=75).
  • The red channel is the expected value (x) minus 3 (25-3=22, 51-3=48, 76-3=73).

  Notice that the value is reduced with a constant value of 4,1,3, independently of the pixel value (so there is a constant effect).
  What I could explain is a pixel value dependable feature, instead of a fixed one.
  What is worse is that if I choose to generate a video with frames consisting in every color (pixel values [255 0 0],[0 255 0] and [0 0 255]) I get the corresponding outputs values ([251 0 0],[0 254 0] and [0 0 252])
  I though that this relation was related to the grayscale Y value, where :

  Y = 76/256 * RED + 150/256 * GREEN + 29/256 * BLUE

  But this coefficients are not related with the output obtained. Maybe the problem is the reading with VideoCapture ?

  EDIT :
  In case that I want to have the same output value for the pixels (Ej : [10,10,10] experimentally I have to create a img where the red and blue channel has the green channel value plus 2 :

  value = 10
  
  img = np.zeros((resolution[0],resolution[1],3),dtype=np.uint8)+value
  
      img[:,:,2]=img[:,:,2]+2
  
      img[:,:,1]=img[:,:,1]+0
  
      img[:,:,0]=img[:,:,0]+2

  Anyone has experience this issue ? It is related to the encoding process or just that OpenCV treats the image differently, prior encoding, depending on the fourcc parameter value ?

• ffmpeg drawtext on video in start

  29 janvier 2017, par Mr Boota

  I’ve to add declaimer in start of video for few seconds, horizontally center with green background, how can I do it with single command and overwrite the existing video without re-encoding