AXELL CORPORATION


SITEMAP Japanese
PRODUCTS
TECHNOLOGY MOVING IMAGE

 Moving image compression 1. The principles of digital moving image compression 2. Motion compensation techniques
  2.1 Half-pixel motion prediction  
  Distortions such as mosquito noise and block noise occur in reference images that have already been decoded, and also have an effect on macroblocks that have already undergone motion compensation using these reference blocks. This type of distortion is called high-frequency distortion, and the use of a low-pass filter is known to eliminate such distortions. Once the reference block is determined through motion prediction, the neighboring block one pixel up or one pixel across can be extracted to create a new block using the average of the two. High-frequency distortions in these blocks are reduced, just like when a low-pass filter is used, and as such further searches can further enhance the quality of the motion compensation. The motion vectors that correspond to these blocks are described in half-pixel units, and thus this scheme is referred to as half-pixel motion prediction.  
  2.2 Inter-coding and intra-coding  
  When the object has deviated from the search window or when it is no longer in the reference image and a new object has appeared, then the difference between the reference block and the macroblock becomes more complex than the macroblock data, and there is a danger that this may increase the code volume. Therefore, when the prediction differential exceeds a given reference value, then this problem can be overcome by using the pixel value of the macroblock and ignoring the difference with the reference block. The process of encoding the pixel value of the macroblock is called intra-coding or intra-frame coding, and the process of encoding the prediction differential after motion compensation is referred to as inter-coding.  
  2.3 GOP  
  In a stream of images, new objects appear and disappear as time goes on, and continual motion compensation will cause a gradual fall in the picture quality of decoded images. Therefore, it is necessary to insert a frame that does not require motion compensation at regular intervals after a fixed number of frames. Also, compression rates can be increased further if up-coming frames can be used as reference images like previous frames are. To achieve this, three types of frames, referred to as pictures, have been defined as follows.  
  ・I-picture (Intra-coded Picture)
All macroblocks are intra-coded.
 
  ・P-picture (Predictive-coded Picture)
Past I- and P-pictures are used as reference images for inter-coding.
 
  ・B-picture (Bidirectional-coded Picture)
Past and up-coming I- and P-pictures are used as reference images for inter-coding.
 
  A stream of images can be broken down into groups of B- and P-pictures preceded by an I-picture (GOP: Group of Pictures) and these groups can be isolated and replayed. One point to note here is that encoding and decoding is performed according to the I, P, B order and so frames must be replaced according to this processing sequence. Fig. 2.3 (a) shows an example of the case where an I-Picture has been inserted after every six frames and then after every three, whereas Fig, 2.3 (b) shows an example of replacement for decoding.  
     
  Fig. 2.3 Order of pictures  
 
 
     
  Further reading
Please refer to the following references for more detailed explanations.
 
PAGE TOP  
 
 Moving image compression 1. The principles of digital moving image compression 2. Motion compensation techniques
 
   

CONTACT TERM  
Copyright(C)2006-2011.AXELL CORPORATION
Axell Corporation