176 IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 3, NO. 3, SEPTEMBER 1999 Wavelet Image Compression—The Quadtree Coding Approach Adrian Munteanu, Jan Cornelis, Member, IEEE, Geert Van der Auwera, and Paul Cristea, Senior Member, IEEE Abstract— Perfect reconstruction, quality scalability, and region-of-interest coding are basic features needed for the image compression schemes used in telemedicine applications. This paper proposes a new wavelet-based embedded compression technique that efficiently exploits the intraband dependencies and uses a quadtree-based approach to encode the significance maps. The algorithm produces a losslessly compressed embedded data stream, supports quality scalability, and permits region- of-interest coding. Moreover, experimental results obtained on various images show that the proposed algorithm provides competitive lossless/lossy compression results. The proposed technique is well suited for telemedicine applications that require fast interactive handling of large image sets, over networks with limited and/or variable bandwidth. Index Terms—Embedded coding, image coding, lifting scheme, lossless compression, medical image compression, progressive image transmission, quadtree coding. I. INTRODUCTION C OMPRESSION methods are important in telemedicine applications for guaranteeing fast interactivity during browsing through large sets of images (e.g. three-dimensional data sets, time sequences of images, image databases), for searching context-dependent detailed image structures, and/or quantitative analysis of the measured data. These applica- tions impose specific constraints in trading off image quality against bit rate and complexity against cost, compression, and decompression speed. In medical imaging, it is not acceptable to lose any in- formation when storing or transmitting an image. There is a broad range of medical image sources, and, for most of them, discarding small image details that might be an indication of pathology could alter a diagnosis, causing severe human and legal consequences [1]. For example, images obtained from projection radiography may reveal lesions by image details that are extremely sensitive to lossy compression since they have poorly defined borders (e.g. some microcalcifications in mammograms, the trabecular pattern of bone, the edge of a pneumothorax, etc.) and are only distinguishable by subtle changes in the contrast [1]. In computer tomography (CT) images, some lesions (e.g., subtle fractures), or the slight changes in density between the gray and white matter of the Manuscript received January 7, 1999; revised March 17, 1999. A. Munteanu, J. Cornelis, and G. Van der Auwera are with the Electronics and Information Processing Department, Vrije Universiteit Brussel, Brussels B-1050, Belgium. P. Cristea is with the Digital Signal Processing Laboratory, Politehnica University of Bucharest, Bucharest 77206, Romania. Publisher Item Identifier S 1089-7771(99)07086-7. brain may be at the limit of the detection device’s spatial and/or contrast resolution, and they risk becoming undetectable due to the effects of lossy compression. Also, subtle changes in the echo texture or some well-defined edges are crucial for diagnostic interpretation of ultrasound images and may be highly sensitive to lossy compression. In other medical applications, like coronary angiography, where one has to measure submillimeter blood vessel diameters at the location of the stenosis, lossy coding methods are obviously inadequate [2]. More generally, for image data, which are considered as measurements and get their interpretation from a quantitative analysis, one usually cannot afford information loss due to compact coding. The current standard for still image compression, JPEG [3], does meet only part of the quality and functionality demands of medical imaging. For instance, JPEG yields a reasonable compression performance at high and intermediate bit rates, but quality is poor for low bit rates. The breakdown at high compression ratios is mainly due to blocking artefacts, caused by the initial partitioning of the image in square blocks within the DCT-based decorrelating module, and the block-based quantization module. With respect to functionality, JPEG supports lossy and lossless compression, but not within the same coded bit stream: we have to select either the classic JPEG coder [3] for lossy coding or the lossless JPEG coder [3], [4] for lossless coding. In wavelet-based coding, no partitioning of the image is required, although the convolutions in the discrete wavelet transform can still be computed efficiently on blocks of data. Hence, the typical blocking artefacts, like the ones occurring in JPEG, are avoided and the computation time is hardly increased. Better quality than JPEG can be obtained at low and intermediate compression ratios, and the ringing artefacts occurring at high compression ratios (mainly in the vicinity of edges and/or in textured regions) are generally less objectionable than the JPEG blocking artefacts. Fast image inspection of large volumes of images transmit- ted over low-bandwidth channels like ISDN, public switched telephones, or satellite networks (traditionally known as tele- radiology) requires compression schemes with progressive transmission capabilities [5]. The ability of the wavelet-based compression techniques to create embedded data streams fa- cilitates the progressive transmission of data over networks with limited and/or variable bandwidth: a coarse version of the image corresponding to a certain refinement level can be transmitted first, decoded, and displayed at the remote site. At this stage, the user can decide either to further refine 1089–7771/99$10.00  1999 IEEE