STANDARD-BASED INFRARED LINE-SCAN IMAGE CODING:

JPEG-2000 vs. MPEG-4 VTC

by D. Milovanovic, B.Wiecek(*), A. Marincic, Z. Barbaric and G. Petrovic

Faculty of Electrical Engineering, P.O.Box 3554, 11001 Belgrade, Yugoslavia Email:

(*)Technical University of Lodz, Institute of Electronics, Stefanowskiego 18/22, 90-924, Poland

1. Introduction

IR remote sensing and reconnaissance applications usually require transmission of line-scan thermal images in real-time. This may impose a severe requirements on the capacity of processing for efficient transmission. A standard way to deal with these problems is to reduce the redundancy of thermal images. Fortunately, statistical properties of IRLS images depend highly on the characteristics of scanner and the effective ground resolution and the correlation between adjacent pixels is usually very high, implaying a high degree of redundancy in the acquired image [1].

Statistical image compression techniques take advantage of the intrinsic features of IRLS images as well as their relation to the final human observer to eliminate redundancy before transmission [2]. Transform coding (TC) has become the de-facto standard technique in image coding (JPEG, MPEG), where the discrete cosine transform (DCT) and wavelets are used to decorrelate pixel values before quantization. However, one the most important IRLS image compression requirements is preservation of the amplitude and phase of the edges as well as good visual quality of the image background. Therefore, it is necessary to carefully select the compression technique and optimize operational parameters in IRLS image coding application [4].

In this paper, we systematized features of advanced transform-based image coding standards. Then, JPEG-2000 and MPEG-4 VTC codecs are compared with respect to compression efficiency, complexity, and functionality in IRLS test image coding.

2. Advanced image coding techniques: comparison methodology

A great effort has been made in research community to deliver advanced image coding standards by providing features inexistent in previous standards, but also by providing higher efficiency for features that exist in others.

JPEGis very well known ISO/ITU-T standard created in the late 1980s. There are several mode defined for JPEG, including baseline, lossless, progressive and hierarchical. Baseline mode is the most popular and supports lossy coding only. It is based on the 8x8 block DCT, zig-zag scanning, uniform scalar quantization and Huffman coding. The lossless mode is based on a predictive scheme and Huffman coding.

JPEG-2000 is a new image compression standard currently being developed by the International Standards Organization (ISO/IEC 15444). JPEG-2000 is based on the discrete wavelet transform (DWT), scalar quantization, context modeling, bit-plane binary arithmetic coding and post-compression rate allocation (Figure1). At first, the discrete transform is applied on the source image data. The transform coefficients are then quantized and entropy coded, before forming the output bitstream. The decoder is the reverse of the encoder.

JPEG-2000 supports a number of functionalities (Table1), many of which are inherent from the algorithm itself. Examples of this is tiling and random access, which is possible because of the independent coding of the code-blocks and the packetized structure of the codestream. Another such functionality is the possibility to encode images with arbitrarily shaped Regions of Interest (ROI). Other supported functionalities are error resilience, multicomponent images, extended pixel precision, compressed domain lossless processing (flipping, rotation,...).

MPEG-4Visual Texture Coding (VTC) is the algorithm used in MPEG-4 standard in order to compress the texture and still images. It is based on the discrete wavelet transform (DWT), scalar quantization, zero-tree coding and arithmetic coding. MPEG-4 VTC supports SNR scalability through the use of different quantization strategies. Resolution scalability is supported by the use of band-by-band scanning (BB), instead of traditional zero-tree scanning, which is also supported. MPEG-4 VTC also supports coding of arbitrary shaped objects, by means of a shape adaptive DWT but does not support lossless coding. Several objects can be encoded separately, possibly at different qualities and then composited at the decoder to obtain the final decoded image.

In this paper we report on compression efficiency of JPEG and MPEG-4 standards in IRLS image coding application. Compression efficiency is simply measured by the achieved compression ratio (CR) (or bit-rate R) for each one of the IRLS test images. As a meassure of coding distortion (D), the root mean square error (RMSE) is used, as well as the corresponding peak signal-to-noise ratio (PSNR). Although RMSE and PSNR are known to not always faithfully represent visual quality, it is the only established, well-known, objective measure that works reasonably well across a wide range of compression ratios. For images encoded with a Region of Interest (ROI) the RMSE, as well as the corresponding PSNR, are calculated both for the ROI and for the entire image.

3. Compression efficiency results

It is well known that the compression performance of an image codec is determined both by the statistical characteristics of the input IRLS image [3] and by the capability of the coding algorithm to explore these characteristics.

In this paper, IRLS images are characterized with 25 images from the test set, covering various types of terrain. The test image IRLS20 (512x512 pixels, 256 gray levels) acquired during night from low altitude aircraft is city area, full of details of various size and contrast (Figure 4).

Using the set of test images, we systematically investigated compression efficiency of software implementations of standard codecs. In order to make fair comparison, we firstly used reference software implementation of codecs to find candidate set of operational parameters and operational range that minimize PSNR coding distortion of IRLS test images for given bit rate. The selected set of operational parameters, we applied in the second step for R&D software (Aware Inc. JPEG2000 codec Version 2.1.5 and e-Vue MPEG-4 VTC codec Release 2.0) and further tune operational parameters maximizing PSNR.

The optimal operational parameters of JPEG and MPEG codecs for given bit-rate 0.1bpp (CR=80) and obtained PSNR values for decoded images are shown in Figure4. Comparative curves PSNR vs. bit-rate in range [0.1-1.50 bpp] (compression ratio [80:1-5.3:1]) for test image IRLS20 are shown in Figure 2. Obviously, JPEG2000 and MPEG-4 VTC codec outperform standard JPEG codec. Considering various test images, we obtained better PSNR values for JPEG2000 codec. Further, we investigated JPEG2000 region of interest (ROI) coding (Figure 3) as well progressive decoding by resolution/quality.

Overall, one can say that JPEG2000 codec is preferred to MPEG-4 VTC in many aspects such as rate-distortion performance, computational complexity and progressive compression of IRLS images. Additionaly, JPEG-2000 standard supports multicomponent/multispectral images with more than 8 bits per pixel.

In our future work, we will investigate perceptual quality of IRLS decoded images and their correlation with PSNR objective measures. Based on this new measures, we will try to further improve adaptive compression efficiency of JPEG2000 codec in tracking spatial variations of statistical characteristics of IRLS images [5].

REFERENCES

[1]Ž.Barbarić, A.Marinčić, G.Petrović, D.Milovanović, "Thermal-image generation by line-scanning technique: a new computer model", Journal Applied Optics: Information Processing, vol. 33, no.14, pp.2883-2890, 1994.

[2]K.R.Rao, Z.Bojkovic, D.Milovanovic, Multimedia communication systems, Prentice-Hall, 2002.

[3]D.Milovanović, A.Marinčić, Ž.Barbarić, G.Petrović, "Statistical analysis of computer generated thermal images based on overall modeling of line-scanning process", QIRT 1994, Sorrento Italy, pp.13-18.

[4]D.Milovanović, B.Wiecek, A.Marinčić, G.Petrović, Ž.Barbarić, "A comparative study of advanced frequency-domain coding techniques in compression of infrared line-scan images", QIRT 1998, Lodz Poland, pp.342-348.

[5]D.Milovanović, A.Marinčić, B.Wiecek, G.Petrović, Ž.Barbarić, "Efficient transform coding of infrared line-scan images based on spatial adaptivity", QIRT 2000, Reims France.

Table1.Functionality matrix (+ indicates that it is supported, the more+ the more efficiently or better it is supported. - indicates that it is not supported).

JPEG / JPEG-2000 / MPEG-4 VTC
Lossless compression performance / +1 / +++ / -
Lossy compression performance / +++ / +++++ / ++++
Progressive bitstreams / ++2 / +++++ / +++
Region Of Interest (ROI) coding / - / +++ / +3
Arbitrary shaped objects / - / - / ++
Random access / - / ++ / -
Low complexity / +++++ / ++ / +
Error resilience / ++ / +++ / +++
Non-iterative rate control / - / +++ / +
Genericity4 / ++ / +++ / ++
1Only using the lossless mode of JPEG. 2Only in the progressive mode of JPEG.
3Tile-based only.
4 Ability to efficiently compress different types of imagery across a wide range of bitrates.

Figure 1. Block diagram of JPEG2000 codec.
Figure2.Decoded test image IRLS20: PSNR vs. BitRate for JPEG, JPEG2000 and MPEG4 VTC codec. / Figure3.JPEG2000 VM2.0 decoded test-image: PSNR vs. BitRate for entire image, background and ROI.

Test image IRLS20
("City"-high altitude 512x512x8) /
JPEG decoded test image (DCT8x8)
BitRate=0.20bpp (CR=40) PSNR=23.86dB

JPEG2000 decoded test image
(I9-7 wavelet, Levels=6, Block=6x6, Quality levels=10)
BitRate=0.20bpp (CR=40) PSNR=25.74dB /
MPEG4 VTC decoded test image
(QualityFactor =47)
BitRate=0.20bpp (CR=40) PSNR=25.47dB

Figure4.Original and JPEG, JPEG2000 and MPEG4 VTC decoded IRLS test image.