Browse Prior Art Database

Extraction of side information by motion analysis and interpolation of the nodes of the motion grid for distributed video coding

IP.com Disclosure Number: IPCOM000168611D
Publication Date: 2008-Mar-18
Document File: 5 page(s) / 918K

Publishing Venue

The IP.com Prior Art Database

Related People

Guillaume Boisson: AUTHOR [+3]

Abstract

The paper is directed to video coding, and more precisely distributed video encoding.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 32% of the total text.

Extraction of side information by motion analysis and interpolation of the nodes of the motion grid for distributed video coding

AUTHORS:

Guillaume Boisson

Patrick Lopez

Dominique Thoreau

The paper is directed to video coding, and more precisely distributed video encoding.

In distributed coding, the decoding consists of using the side information (Y) available at the decoder to correctly decode the information sent in a distributed manner (X). To achieve this, channel coding methods are traditionally used, Y being considered as a noisy version of X, a version corrupted by a virtual ‘channel’.  Figure 1, below depicts a block diagram of a distributed coding system without feedback. 

Figure 1: distributed coding system (without feedback)

The present teachings regard the extraction of side information from data available to the decoder. This step is crucial for the efficiency for the distributed coding system.

Typically, a motion estimator is used at the decoder.

The frames of the video flow are divided into two categories:

  • the key-frames (sent in a conventional manner),
  • and the frames to send by distributed coding (DC-frames).

This concept is depicted in Figure 2 which illustrates key-frames and distributed coding frames. 

Figure 2: key-frames and distributed coding frames

In some diagrams of prior teachings, the motion estimation is carried out between the local (and partial) decoded data of the current DC frame and the previous frame (or key-frame), after transmission of a certain quantity of information (hash-based motion estimation [1]).

In other diagrams of prior teachings, the motion estimation is carried out between two successive key-frames and the DC-frame is predicted by interpolation.

Some prior literature proposes motion description by block and by meshing [2]. Only the translational model (2 parameters) is cited.

In the diagrams having recourse to the interpolation – especially the block diagrams –, the current frame is predicted by duplicating the motion information co-located in the current key-frame (see Figure 3): the grid of the current key-frame is reproduced on the DC-frame to predict, and only the translation vector is interpolated. The coefficients of the interpolation depend on the distance between the current frame and the key-frames.

Figure 3: Estimation of backward motion and co-located interpolation (N=2)

The distributed coding diagrams of the prior art literature are confined to descriptions of translation motion (by block on regular grid or by meshing) such as used in the standard encoders, whereas the constraint of concision of the motion information no longer exists (the information is no longer sent).

In addition, the principle of hash-based methods is unsatisfactory. Indeed, these methods rely on the conventional transmission of information (non-distributed) as a prea...