Pattern Recognition 39 (2006) 1492 – 1508 www.elsevier.com/locate/patcog A new algorithm for the embedding of a prediction mechanism into the JPEG2000 coding chain Marco Aguzzi ∗ , Maria Grazia Albanesi, Marco Martelli Dipartimento di Informatica e Sistemistica, Università degli Studi di Pavia, Via Ferrata 1, 27100 Pavia, Italy Received 20 July 2005; received in revised form 8 February 2006; accepted 10 February 2006 Abstract In this paper, we propose the embedding of a prediction mechanism into a part of the coding structure of JPEG2000 image compression standard, in order to reduce the amount of bits sent to the arithmetic coder, without any significant changes into the standard architecture and without loosing performance. The prediction is based upon an innovative processing of the data structures used by the standard JPEG2000 in progressive coding and the addition of a Prediction Matrix, whose computation does not add any overhead at the decoder side. Experiments are performed to test the efficacy of the prediction mechanism, and results are compared to the standard JPEG2000 and other similar approaches. Tests are documented over a set of well-known images from literature, also against different kinds of added noise. Performance, in terms of saved bits are reported, and a new figure of merit is defined to test the efficiency of Prediction. The results prove that the new proposal overcomes the standard and other related approaches for the entire set of referenced images, with significant gain in synthetic images, also in presence of noise.  2006 Pattern Recognition Society. Published by Elsevier Ltd. All rights reserved. Keywords: JPEG2000; Prediction; Wavelet transform; Image compression 1. Introduction With the increasing use of digital media, both in multime- dia processing and transmission over networks, the efficient storing and compression of data is a key issue in the current research fields. In particular, the new JPEG2000 [1–5] stan- dard for still image coding is a promising framework, as it overcomes some limitations of the existing standards, such as JPEG [6–8]. In fact, thanks to the features of its encoding scheme EBCOT [9], the use of multiresolution of wavelet decomposition, the context-based arithmetic coding and the rate-distortion optimization, the new standard is able to as- sure several interesting features: (a) lossy, visually lossless and lossless compression with a better image quality (both from an objective and a subjective point of view), (b) pro- gressive coding and decoding, (c) region of interest defini- tion, and (d) robustness to bit errors. However, despite of ∗ Corresponding author. Tel.: +39 0382985523; fax: +39 0382525638. E-mail address: marco.aguzzi@unipv.it (M. Aguzzi) URL: http://orfeo.unipv.it/aguzzi. 0031-3203/$30.00  2006 Pattern Recognition Society. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.patcog.2006.02.015 its innovative features and high performance, research activ- ities in the field of still image compression are far from be- ing concluded. In particular, new efforts are currently under development [10]: the study of efficient wavelet coefficient modeling [11], the investigation of a useful application- based definition of ROIs [12–14], the analysis of the contexts to enhance the arithmetic coder [15]. In this paper, we investigate the problem of reducing the amount of bits produced by some phases of the stream generation with the following aims: (a) without affecting the global archi- tecture of the standard, (b) without loosing compression performance, and (c) without dramatically increasing the complexity of the decoder. Our approach consists of embedding a Prediction mech- anism in the JPEG2000 coding step called cleanup pass. Prediction is a well-known idea in image coding and trans- mission; in our case, the prediction is performed in such a way that the prediction error is always kept to zero and the complexity of the realization is kept low, as it requires only the construction of a Prediction Matrix (linear with the number of the coefficients) from the coder point of view,