| Summary: | 碩士 === 國立中正大學 === 資訊工程學系 === 85 === Digital images are more and more popular in our daily life and
the demand of digital images has grown in the recent years. But
the data sizes of digitalimages are usually large. In order to
save the storage space and to reducethe transmission time,
several techniques for image compression were developed.However,
many annoying artifacts will appear in an image when the image
is codedat lower bit rates. For example, the blocking effects
will be visible along the block boundaries, the ringing effects
will appear around the sharp edges,and the staircase noise will
arise on the skew edges. Traditionally, low-pass filters are
used to reduce blocking effects. Though the blocking effectsare
reduced by low-pass filters, the details of the images such as
edges aredegraded at the same time. In this thesis, a new post-
processing approach isproposed to reduce blocking effects and
preserve image details.Our approach consists of three main
procedures, segmentation, filtering withAR model coefficients,
and 1-D filtering. The input image is first segmentedinto
several non-overlapping regions. The pixels within the same
region havesimilar gray levels and the edges in the original
image appear as region boundaries. Auto=regressive (AR) model
coefficients for each region are extracted by the two-
dimensional linear analysis. The model coefficientsfor a region
can reflect the characteristics of this region. Therefore, the
filter with these model coefficients is then adopted to reduce
the blockingeffects and preserve the characteristics within this
region. The above procedures work iteratively to get better
performance. This iterative methodcan also reduce the ringing
effects around the sharp edges. At last, 1-Dfilters along the
skew edges are used to reduce the staircase noise.The proposed
approach is tested on several images. The results are
comparedwith the low-pass filter by the peak signal-to-noise
(PSNR) ratio criterion.The experimental results demonstrate that
a significant improvement in subjective and objective quality is
achieved. This shows the feasibility ofthe proposed approach.
|
|---|
