Browse Prior Art Database

NEW ALGORITHM FOR HALF PIXEL MOTION ESTIMATION

IP.com Disclosure Number: IPCOM000009450D
Original Publication Date: 1999-Sep-01
Included in the Prior Art Database: 2002-Aug-26
Document File: 4 page(s) / 162K

Publishing Venue

Motorola

Related People

Manish Singhal: AUTHOR [+2]

Abstract

The H.263 recommendation of ITU-T for Video Coding for Low Bit Rate Communication uses the Half Pixel Motion Estimation Technique to remove temporal redundancy in the picture. The Half Pixel Motion Estimation gives good compression which is essential for low bit rate communication but at the same time is very computation intensive. In this report, we present a new technique which results in a reduction in the computational requirements.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 48% of the total text.

Page 1 of 4

Developments Technical 0 M MOIOROLA

NEW ALGORITHM FOR HALF PIXEL MOTION ESTIMATION

by Manish Singhal and Bhaskar Bhupalam

ABSTRACT

  The H.263 recommendation of ITU-T for Video Coding for Low Bit Rate Communication uses the Half Pixel Motion Estimation Technique to remove temporal redundancy in the picture. The Half Pixel Motion Estimation gives good compression which is essential for low bit rate communication but at the same time is very computation intensive. In this report, we present a new technique which results in a reduction in the computational requirements.

BACKGROUND

  In Video Coding standards like H.261, MPEGl etc., motion estimation is done on the integer pixel boundaries only. However, H.263 goes a step fur- ther and does the motion estimation on the sub pixel positions also, which gives additional compression. A brief description of integer and half pixel motion estimation is given below.

  This will be explained with reference to Figures 1 and 2. Figure 1 shows a typical picture frame with the various Macro Blocks, and Figure 2 shows an enlarged view of the dotted area.

Legend

Macro block in discussion

I I I-J Area shown enlarge in Fig. 2

Fig. 1 Frame Diagram showing Macro blocks

0 hwm.da, 1°C. 1999 90 Sepiember 1999

[This page contains 15 pictures or other non-text objects]

Page 2 of 4

Developments Technical 0 M MOTOROLA

Legend

- Macro Block Boundary

- Intreger Pixel Boundary

- Half Pixel Boundary

X Intreger Pixel Position

0 Half Pixel Position

Fig. 2 Macro block Diagram showing pixels and half pixels

DESCRIPTION OF THE INTEGER AND In the half pixel case, the value of the half pixels HALF PIXEL MOTION ESTIMATION is calculated by bilinear interpolation of the integer pixels given by the following equations:

  
In integer motion estimation, the selected macro
block is moved around in the previous frame at
pixel boundaries. So, the macro block would take
positions like A, B, C etc. But in the case of half
pixel motion estimation, the positions like b, c, d
etc. (shown by '0' in Figure 2) are also taken into
account.

b = (A+B)/2

c = (A+C)/2

d = (A+B+C+D)/4

91 September 1999

[This page contains 15 pictures or other non-text objects]

Page 3 of 4

Developments Technical 0 M MOTOROLA

BMotomla.lnc. I999 92 September 1999

  Usually, first the integer pixel motion estimation is done after which the half pixel motion estimation around the estimated integer motion vector is done. For example, let the motion vector for the macro block in Figure 2 is (1, -1) i.e. position A. This would mean that for estimating the half pixel motion vector, the estimation would be done around posi- tion A. These positions are 1,2...8 as shown in Figure 3. Typical implementations have the follow- ing order:

  1. Interpolate the total image used for predic- tion.

  2. Estimate integer motion vectors for all macro blocks.

  3. Estimate half-pixel motion vectors for all the macro blocks. Follow the order 1,2,3,4,5,6,7, 8 in this process.

Fig. 3 Half pixel positions a...