Professional Documents
Culture Documents
2 FEBRUARY 2007
606
LETTER
A Low-Complexity Interpolation Method for Deinterlacing
Copyright
c 2007 The Institute of Electronics, Information and Communication Engineers
LETTER
607
x(i, j − 1), if Dh = 0
− 1, j − 1) + x(i + 1, j) +
(x(i 3. Simulation Results
x(i − 1, j) + x(i + 1, j + 1))/4,
if min(D d1 , Dd2 , Dv ) = Dd1 For an M × N image, we remove every odd vertical line
x(i, j) =
(x(i − 1, j) + x(i + 1, j − 1) + (2) and thus obtain the M/2 × N image. Then, we reconstruct
x(i − 1, j + 1) + x(i + 1, j))/4, the M × N image by interpolating the odd lines for differ-
if min(D , Dd2 , Dv ) = Dd2 ent methods. Totally, ten 512 × 512 standard test images
d1
(x(i − 1, j) + x(i + 1, j))/2,
if min(D , D , D ) = D with 8-bit brightness resolution are considered. The results
d1 d2 v v of peak signal-to-noise ratio (PSNR) are illustrated in Ta-
ble 1. Except our method, five previous interpolation meth-
Obviously, the proposed method requires only simple in-
ods, FOI [2], ELA [3], ELC [4], APM [5] and MBI [6], are
teger operations, such as addition, subtraction and abso-
used for comparisons. It can be observed from the results
lute value, so it can be easily implemented with hard-
that our algorithm achieves better quantitative quality than
ware. The simulation results shown in the following section
the previous interpolation methods. Certainly, the exact de-
will demonstrate that our method achieves excellent perfor-
gree of improvement is dependent on the content of different
mances in both objective and subjective image quality.
images being processed.
In addition, three well-known 8-bit gray-level video se-
quences, Football, Akiyo and Erik, are used to test various
interpolation methods. The size of each frame is 352 × 288,
and the frame rate is 30 per second. Each individual frame [1] G.D. Haan and E.B. Bellers, “Deinterlacing–an overview,” Proc.
is down sample by a factor of 2 vertically, and then inter- IEEE, vol.86, no.9, pp.1839–1857, Sept. 1998.
polated by using different approaches. Tables 2 and 3 show [2] R.S. Prodan, “Multidimensional digital signal processing for televi-
the average and worst-case PSNR of the first 50 frames for sion scan conversion,” Phillips J. Res., vol.41, pp.576–603, 1986.
[3] T. Doyle, “Interlaced to sequential conversion for EDTV applica-
the video sequences respectively. The proposed method tions,” Proc. 2nd Int. Workshop on Signal Processing of HDTV,
consistently produces better performance. This demon- pp.412–430, Feb. 1988.
strates the capability of our method in coping with video [4] T. Chen, H. Wu, and Z.H. Yu, “Efficient de-interlacing algorithm
sequences. using edge-based line average interpolation,” Optical Engineering,
For subjective testing, the interpolated Lena images vol.39, no.8, pp.2101–2105, Aug. 2000.
[5] H.-C. Kim, B.-H. Kwon, and M.-R. Choi, “An image interpolator
using different interpolation methods are shown in Fig. 2.
with image improvement for LCD controller,” IEEE Trans. Consum.
A portion of each image is enlarged in order to evaluate the Electron., vol.47, no.2, pp.263–271, May 2001.
visual quality. FOI produces noticeable artifacts in areas of [6] M.Q. Phu, P.E. Tischer, and H.R. Wu, “A median based inter-
angled edges as shown Fig. 2 (b). Like those previous pro- polation algorithm for deinterlacing,” Proc. International Sympo-
posed methods [4]–[6], our method shows visually pleasing sium on Intelligent Signal Processing and Communication Systems,
picture especially at image edges. pp.390–397, 2004.
[7] D. Wang, A. Vincent, and P. Blanchfield, “Hybrid de-interlacing al-
To explore the computational complexity, we list the gorithm based on motion vector reliability,” IEEE Trans. Circuits
operations required to interpolate each missing pixel for Syst. Video Technol., vol.15, no.8, pp.1019–1025, Aug. 2005.
various interpolation methods in Table 4. Compared with [8] X. Gao, J. Gu, and J. Li, “De-interlacing algorithms based on mo-
the newly interpolation methods [4]–[6], our method re- tion compensation,” IEEE Trans. Consum. Electron., vol.51, no.2,
quires lower computational complexity (5 absolute, 5 addi- pp.589–599, May 2005.
tion and 11 subtraction operations). Although the proposed [9] S. Yang, Y.Y. Jung, Y.H. Lee, and R.H. Park, “Motion compensa-
tion assisted motion adaptive interlaced-to-progressive conversion,”
method needs more operations than both FOI and ELA do, it IEEE Trans. Circuits Syst. Video Technol., vol.14, no.9, pp.1138–
achieves better objective image quality as shown in Tables 1, 1148, Sept. 2004.
2, and 3. [10] O. Kwon, K. Sohn, and C. Lee, “Deinterlacing using directional in-
terpolation and motion compensation,” IEEE Trans. Consum. Elec-
tron., vol.49, no.1, pp.198–203, Feb. 2003.