anbul 2004
QUANTITATIVE ANALYSIS OF IMAGE QUALITY
OF LOSSY COMPRESSION IMAGES
Ryuji Matsuoka*, Mitsuo Sone, Kiyonari Fukue, Kohei Cho, Haruhisa Shimoda
Tokai University Research & Information Center
2-28-4 Tomigaya, Shibuya-ku, Tokyo 151-0063, JAPAN
(ryuji, sone3)@yoyogi.ycc.u-tokai.ac.jp, (fku, kcho, smd)@keyaki.cc.u-tokai.ac.jp
KEY WORDS: Image, Compression, Quality, Analysis, Experiment, Texture, Color
ABSTRACT:
High resolution images acquired by an aerial digital camera and high resolution satellite images are expected to become more
powerful data source of GIS. Since the large data volume of a high resolution image brings difficulties in dealing with it, lossy
image compression is going to be indispensable. Image quality of a reconstructed image after decompression is usually evaluated by
visual inspection. Although some numerical measures such as RMSE or PSNR are used to compare various image compression
techniques, numerical evaluation of quality of a reconstructed image is seldom conducted. Therefore, we decided to carry out an
empirical investigation into the effects of lossy image compression on quality of color aerial images by using color and texture
measures. From the experiment results, it can be concluded that color space conversion and downsampling in JPEG compression
have an effect on quality of a reconstructed image. The-results supported that lossy JPEG 2000 compression is superior to lossy
JPEG compression in color features. However, lossy JPEG 2000 compression does not necessarily provide an image of good quality
in texture features. Moreover, the results indicated that an image of finer texture features is less compressible, and quality of the
reconstructed image is worse in both color and texture features. Finally, it was confirmed that it is difficult to set an appropriate the
quality factor, because the optimal setting of the quality factor varies from one image to another.
1. INTRODUCTION compression on quality of a reconstructed image by using
numerical measures of image quality.
High resolution images acquired by an aerial digital camera and
high resolution satellite images such as IKONOS images are
expected to become more powerful data source of GIS. 2. LOSSY IMAGE COMPRESSION
Resolution of images for urban GIS is usually desired to be as
high as possible. The higher resolution of the image is, the 2.1 JPEG Compression
larger its data volume is. The large data volume of a high
resolution image brings difficulties in dealing with it. Therefore, From the point of view of interoperability, lossy JPEG
image compression is going to be required. Since the compression and lossy JPEG 2000 compression are desirable
compression ratio achieved by lossless image compression is compressions at the moment.
unsatisfactory for this purpose, lossy image compression is
indispensable. Lossy JPEG compression based on the discrete cosine
transform (DCT) is the past and current still image compression
Quality of a reconstructed image after decompression is usually standard. On the other hand, lossy JPEG 2000 compression
evaluated by visual inspection. Although some numerical based on the discrete wavelet transform (DWT) is the current
measures such as RMSE (root mean square error) and PSNR and future still image compression standard. However, JPEG
(peak signal to noise ratio) are used to compare various image 2000 compression has not yet come into wide use. Accordingly,
compression techniques, numerical evaluation of quality of a we decided that the main target of the study was lossy JPEG
reconstructed image is seldom conducted. Therefore, we compression.
carried out an empirical investigation into the effect of lossy
Source > M. Entro
; Source Color Space 1p! Downsampling =» Forward DCT [=| Quantization =%» ©
image data conversion encoding
| JPEG compression |
[ JPEG decompression ] |
Xv
; "m" + ence iu Entro
Decompressed Color space Upsampling |« Inverse DCT. 44 Dequantization |« TIN
image data conversion g
Figure l. JPEG compression and decompression process flow
647
