用GDAL对影像重采样的一些要点

1、RasterIO只能使用最临近插值法

做影像金字塔时，开始用的GDALDataSet的RasterIO通过控制行列宽度来抽层，代码写起来很简洁，奇怪一直没见设置重采样参数的地方，急于实现功能，没有怎么深究，基本功能全都做好后发现分层质量很差。文档里边没找到资料介绍这些，查了很久在Maillist上发现RasterIO() level resampling should all be "nearest neighbour", and so

not introduce any new values.看来不换方法是不行了。

参考资源：http://lists.maptools.org/pipermail/gdal-dev/2005-May/005646.html

On 5/4/05, Sheykhet, Rostic <rsheykhet at sanz.com> wrote:

> Hello,

>

> I am trying to perform RasterIO on a band that references a color

table.

> When the source and destination image dimensions are different, I get

> back a band that has indexes into blank/invalid color table entries.

> I assume this is because of the decimation procedure. Is there a way

> to modify this behavior?

Rostic,

RasterIO() level resampling should all be "nearest neighbour", and so

not introduce any new values. Is it possible you have built overviews

on the file in question with averaging? Perhaps the format is a lossy

compression format?

Best regards,

--

---------------------------------------+--------------------------------

---------------------------------------+------

I set the clouds in motion - turn up | Frank Warmerdam,

warmerdam at pobox.com

light and sound - activate the windows | http://pobox.com/~warmerdam

and watch the world go round - Rush | Geospatial Programmer for Rent

2、 GDALWarp的四种采样方式

enum GDALResampleAlg

Warp Resampling Algorithm

Enumerator:

GRA_NearestNeighbour		Nearest neighbour (select on one input pixel)
GRA_Bilinear		Bilinear (2x2 kernel)
GRA_Cubic		Cubic Convolution Approximation (4x4 kernel)
GRA_CubicSpline		Cubic B-Spline Approximation (4x4 kernel)

GDALWarpOpations::eResampleAlg

One of GRA_NearestNeighbour (the default, and fastest), GRA_Bilinear (2x2 bilinear resampling) or GRA_Cubic. The GRA_NearestNeighbour type should generally be used for thematic or colormapped images. The other resampling types may give better results for thematic images, especially when substantially changing resolution.

3、用GDAL Utility实现高质量重采样

===============================================================
NiceResample.cmd
===============================================================
@echo off
echo.
echo.   %~n0: resample an image nicely, slow but with good results
echo.
if [%2]==[] goto :Usage
if not exist "%1" goto :NoFile
rem if exist "%2" goto :OutExists
set inFile=%1
set outFile=%2

:: enables variable expansion inside FOR loop (!x! syntax)
:: requires Windows 2000 or newer
setlocal enabledelayedexpansion

:Main
        set tmpFile=%infile%
        :: resample from 900x900 pixels down to 128x128
        :: in 100px stages.
        for %%a in (900 800 700 600 500 400 300 200 128) do (
                gdalwarp -q -ts %%a %%a !tmpFile! xx_%%a_%inFile%
                set tmpFile=xx_%%a_%inFile%
                )
        :: "&& del" only deletes if previous command successful
        gdal_translate !tmpFile! %outFile% && del xx_*_%inFile%
        echo.
        echo   Finished writing 128x128px %outFile%
        goto :EOF

:NoFile
        echo *** Error: input %1 not found
        goto :EOF

:Usage
        echo. Usage: %~n0 [input image] [output image]
        echo.
        echo   output overwritten if it exists
        goto :EOF
===============================================================

参考：Resampling With GDAL库

http://www.nabble.com/Resampling-With-GDAL-t3611789.html

4、使用调色板的影像只能使用nearest neighbour采样方式。

For indexed images change the resampling to nearest neighbour (-rn)
rather than bilinear, cubic or cubic spline (-rb, -rc, -rcs):

参考：http://www.nabble.com/Resampling-With-GDAL-t3611789.html