图像快速缩放算法
4星 · 超过85%的资源 需积分: 16 133 浏览量
2013-04-07
14:12:29
上传
评论
收藏 244KB PDF 举报 
A FAST METHOD FOR SCALING COLOR IMAGES
Jaana Parkkinen, Mikko Haukijärvi, Petri Nenonen
Nokia Corporation
P.O. Box 1000, FI-33721 Tampere
Finland
Jaana.Parkkinen@nokia.com
, Mikko.Haukijarvi@nokia.com, Petri.Nenonen@nokia.com
ABSTRACT
Image scaling is an important processing step in any
digital imaging chain containing a camera sensor and a
display. Although resolutions of the mobile displays have
increased to the level that is usable for imaging, images
often have larger resolution than the mobile displays. In
this paper we propose a software based fast and good
quality image scaling procedure, which is suitable for
mobile implementations. We describe our method in
detail and we present experiments showing the
performances of our approach on real images.
KEY WORDS
Mobile phone, camera, display, image scaling
1. Introduction
Over 5 megapixel size sensors are typical to digital
cameras also in the imaging phones, and the sizes are
constantly increasing. The display sizes have not
increased at the same pace on the mobile devices.
Therefore the captured image size has to be reduced so
that it fits into the display. This means image downscaling
using decimation methods. Sometimes an image is
smaller than the display, or it contains an interesting
detail. In this case the image size can be enlarged. The
zooming requires upscaling using interpolation methods.
The basic decimation and interpolation methods are very
low in complexity and effectively implementable, but
produce severe aliasing and pixelization artifacts. The
downscaling and upscaling algorithms have to have an
adequate quality. Otherwise artifacts, such as aliasing
effects, jagged edges, excessive smoothing or
pixelization, are introduced to images.
Mobile platforms set strict limits to the amount of
memory and processing power available for image
processing and enhancement algorithms. Large images
consume lot of memory and processing power. The
amount is directly or exponentially relative to the number
of pixels in an image.
The sampling of signal is an important part of signal
processing theory, and it is widely covered in the
literature. [1] There are presented several possibilities to
do image downscaling and upscaling in the literature of
the signal and image processing. [2,3] In downscaling
many input pixels correspond to one output pixel and in
upscaling vice versa. In the basic downscaling method
only one of the input pixels is selected to be the output
pixel. This is called the nearest neighbor method. The
nearest neighbor method produces severe aliasing
artifacts. The common downscaling methods include
antialias filter and re-sampling. The downscaled data are
most often taken as a linear combination of the sampled
input data and a certain kernel.
Sometimes only a part of the image contains interesting
information. Varying level of zooming with panning
support is required for showing the details at the area of
interest. The zooming can be implemented using
upscaling algorithms. The basic upscaling method is
called pixel copy, which means copying one input pixel to
multiple output pixels. This method causes pixelization
and blocking artifacts. Better results are achieved by
using more advanced methods that use some spatial
filtering. There exist many methods with different
complexity.
Because of varying sizes of source and target images
methods supporting all possible scaling ratios are needed.
Bilinear interpolation is a generally known method. In
this method the output pixel is a weighted average of the
nearest input pixels. The weights can be computed
effectively for any scaling ratio. Therefore the bilinear
interpolation is a good compromise between complexity
and quality. A weighted average of the input pixels can be
used also in the decimation case.
In this paper we introduce a novel and computationally
effective solution to the problem of scaling of digital
images. Our proposed method is a LUT (Look-up Table)
based weighted average processing and it is fast and
suitable for mobile implementations. We provide detailed
description of the algorithm.
17th European Signal Processing Conference (EUSIPCO 2009) Glasgow, Scotland, August 24-28, 2009
© EURASIP, 2009 2032
资源评论
yemuzi2014-07-16不错，很有参考价值
不思进步2014-02-28英文的文档，没有代码
hallucaQQ2013-11-18有启示，谢谢
ardeth2013-12-06谢谢大大，这个很有用，正是我想找的
