An Efficient Method for DCT-Domain Image Resizing with Mixed Field/Frame-Mode Macroblocks

1

An Efficient Method for DCT-Domain Image Resizing with Mixed Field/Frame-Mode Macroblocks

Changhoon Yim and Michael A. Isnardi

IEEE TRANSACTION ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 9, NO. 5, AUGUST 1999

2

Outline

Introduction DCT-Domain Image Resizing Decomposition of DCT-Domain Image

Resizing Operation Proposed DCT-Domain Image Resizing

Method Discussion on DCT-Domain Filtering Experimental Result Conclusion

3

Introduction

Many image and video compression standards such as JPEG, MPEG-1, MPEG-2, and H.261 are based on DCT coding techniques.

In MPEG-2, there are field and frame pictures to accommodate coding of progressive and interlaced video.

Previous DCT-domain image resizing methods assume that all macroblocks are in the frame mode.

4

Field and frame mode of a macroblock in the spatial domain

5

DCT-Domain Image Resizing : Four Steps Low-pass filtering in the DCT domain Inverse DCT 2 : 1 downsampling Forward DCT

6

DCT-Domain Image Resizing : Matrices

WC

CCST

1

1

0

0

1616:

168:

88:

W

S

C

01

01

01

018 IS

DCT transform matrix

Downsampling matrix

Low-pass filter

00

042

IIW

tTXTY

7

Decomposition of DCT-Domain Image Resizing Operation

tt

t

t

CSC

CWXW

C

CCSY

0

0

0

0

ts

tcs

wt

t

c

w

CCXY

SSXX

C

CX

C

CX

WXWX

0

0

0

0

WW

CCt

t

1

0000

00

0000

00

8

Inverse DCT for a DCT block

jijit

tt

tt

wt

t

w

XCXC

CXCCXC

CXCCXC

C

CX

C

C

XX

XXX

,,

1,10,1

1,00,0

1,10.1

1,00,0

of IDCT thesreprestent

0

0

0

0

9

Downsampling in the spatial domain

10

Proposed DCT-Domain Image Resizing Method

11

Frame- and Field-mode resizing transformation matrices

12

Discussion on DCT-Domain Filtering (1) Previous methods are simple low-pass filtering by tr

uncation in DCT domain The matrix is not equivalent to sharp low-pass filter i

n DFT domain, and dose not correspond to a liner convolution in the spatial domain

May produce visible aliasing and blocking artifacts Affect differently the field- and frame-mode in spatial

domain may result in serious blocking artifacts between field- and frame-mode macroblocks

13

Discussion on DCT-Domain Filtering (2) The implicit low-pass filtering operation as av

eraging by the downsampling matrices would be a good compromise for block-based DCT-domain image resizing

Would not have any blocking artifact The filtering characteristics should be the sa

me for field- and frame-mode macroblocks

14

Experimental Results (1)

15

Experimental Results (2)

16

Experimental Results (3)

17

Conclusion

Propose method can handle field-mode Avoid the blocking artifacts and the difference

in the characteristics between field- and frame-mode macroblocks

Future research could be performed for more efficient for low-pass filtering in the DCT domain for DCT-domain image resizing