AN EFFECTIVE DE-INTERACING TECHNIQUE USING
MOTION COMPENSATED INTERPOLATION
IEEE TRANSACTION ON Consumer Electronics, AUG. 2000
You-Young Jung, Byung-Tae Choi, Yung-Jun Park and Sung-Jea Ko
2005. 05. 03
Ki-Hun Han
Sejong University, DMS Lab.
< Abstract >
 A new de-interlacing algorithm using motion compensated
interpolation is suggested
 First, motion estimation is performed between the same parity
fields
 Then the motion vector is refined in the interpolation field
 For further improvement, a five point median filter with temporal
emphasis is used to reduce the interpolation error caused by
incorrect motion
Sejong University, DMS Lab.
< ⅠINTRODUCTION >
 Current TV systems has visual artifacts such as an edge flicker, an
interline flicker and line crawling
 De-Interlacing which is a picture format conversion form
interlaced to progressive picture has been widely used to reduce
the those artifacts
 HDTV systems support the progressive scan but camera and TV
system still support the interlaced scan
 Various de-interlacing techniques
fixed vertical-temporal filtering[2], edge-based line average[4]-[5],
motion adaptive techniques[6]-[7]
Sejong University, DMS Lab.
< ⅠINTRODUCTION >
 The basic concept of motion adaptive schemes is to select the
appropriate interpolation method according to the motion of the
image – spatial filtering in motion area and temporal filtering in
static area
 But, this approach is highly dependent on the accuracy of motion
estimation
 We propose a new de-interlacing scheme using the motioncompensated field interpolation and the field merging technique
Sejong University, DMS Lab.
< ⅡDE-INTERLACING TECHNIQUES >
The ELA-based technique is the most popular since it exhibit good
performance with small computational load
The ELA algorithm uses the directional correlations between pixels to
interpolate
Sejong University, DMS Lab.
 ELA-based technique introduces flicker at
motion area
 One of solutions is motion compensation
for inter-field processing
 Before motion estimation, the previously
de-interlaced field is inserted in the current
missing field
 Motion Estimation
 Three-point median filter
Sejong University, DMS Lab.
Ⅲ PROPOSED MOTION COMPESATED DE-INTERACING
fn-1, fn and fn+1 denote the previous, current and next field, respectively
Sejong University, DMS Lab.
Ⅲ-A The Motion Compensated Field Interplation
Sejong University, DMS Lab.
Ⅲ-A The Motion Compensated Field Interplation
 It is often observed that inconsistencies or non-smoothness in the
estimated vector field decreases the interpolated picture quality
 We present the proposed smoothing scheme: Let B and Ni, where
i=1,2,…,8 denote the current block and the eight nearest neighboring blocks
around B in the interpolated field
Sejong University, DMS Lab.
Ⅳ FURTHER IMPROVEMENTS
 The performance of the MCFI is satisfactory if the motion vectors are accurate
and reliable. However, when the estimated motion vector is incorrect, the
interpolated picture quality can be degraded
 In this case, the intra-field interpolation using spatial information might be more
reliable
 The proposed scheme uses a five-point median filter for intra and inter-field
interpolation
Sejong University, DMS Lab.
Ⅴ EXPERIMENTAL RESULTS
 For motion estimation, the block size is 16x8, search range for initial
motion estimation is ±8, and the search range for refinement of the initial
motion vector is ±2
Sejong University, DMS Lab.
Ⅴ EXPERIMENTAL RESULTS
Sejong University, DMS Lab.
Ⅴ EXPERIMENTAL RESULTS
Performance of de-interlacing methods on Flower garden sequence
Sejong University, DMS Lab.