Compression of map images for real-time applications

Abstract

Digital maps can be stored and distributed electronically using compressed raster image formats. We introduce a storage system for the map images that supports compact storage size, decompression of partial image, and smooth transitions between various scales. The main objective of the proposed storage system is to provide map images for real-time applications that use portable devices with low memory and computing resources. Compact storage size is achieved by dividing the maps into binary layers, which are compressed using context-based statistical modeling and arithmetic coding. Partial image decompression is supported by tiling the image into blocks and implementing direct access to the compressed blocks. In this paper, we give overview of the system architecture, describe the compression technique, and discuss implementation aspects. Experimental results are given both in terms of compression ratios and image retrieval timings.

Introduction

Real-time cartography imaging application provides user with the view of geographic map for the area surrounding the user's location [1], [2]. The system may use global positioning service (GPS) [3] or mobile positioning service (MPS) [4] for obtaining the coordinates of the current location. The location can be updated in real-time (about once or twice in every second). The system must also support real-time panning (spatial movement) and zooming (change of resolution) on the map. By panning, we mean scrolling the map; and by zooming, we mean the change of the view on the display in a closer or wider perspective.

Digital maps are usually obtained from spatial databases [5], [6] where the maps are stored in vector formats. The visual outlook of maps representing the same region varies depending on the type of the map (topographic or road map), and on the desired scale (local and regional maps). Individual map images are reproduced for each scale separately and stored as separate raster images augmented with the location information of the map. A typical map image needs only a few color tones but high spatial resolution for representing the details such as roads, infrastructure and names of the places.

In on-line map imaging applications, the images are usually stored in an inefficient, uncompressed raster form. The storage size of a map image is huge. For example, electronic library of Finnish road maps of the resolution 1:250,000 takes an entire CD (over 600 Mb) in uncompressed form [7]. In comparison, the portable viewing device, such as pocket computers, have typically about 64 Mb of the storage space, which can be expanded at present by about 256 Mb through using compact flash memory cards [8]. The storage requirements of the maps can therefore be a bottleneck, especially in the case of portable devices, in which the maps share the limited memory resources with the operating system, application and other data.

A better approach is to provide the user with the images in compressed form [9], [10]. For example, an uncompressed black-and-white topographic image of 5000×5000 pixels takes about 3 Mb in uncompressed form. The latest compression standards [11], however, can compress typical map images by a factor of about 20:1, which corresponds to the file size of 150 kb. A drawback of the existing compression techniques is that the entire image must be decompressed in memory before the image can be viewed. This can be a problem if the device does not have sufficient computing resources for real-time image decompression.

In this paper, we propose map image storage system (further denoted as MISS), in which we present reasonable solutions both to the storage problem and to the real-time requirements of the system. The MISS images are composed of semantic binary layers, which are compressed using a context-based statistical modeling and arithmetic coding as shown in Fig. 1. The method is basically the same as in the latest international compression standards, Joint Bi-level Image Group (JBIG) and JBIG2 [12], [13], [14] with a few differences described later.

To meet the real-time requirements, we provide direct access to the compressed image file. Our approach is to divide the image into b×b non-overlapping rectangular blocks, which are compressed separately. The compressed blocks are stored in the same file, and an index table is stored in the header of the file to locate the starting points of the code blocks. In this way, direct access can be provided with the accuracy of the block size. The block size is a compromise between compression efficiency and the decoding speed. The JBIG2 file structure supports this kind of file organization, where the image is composed of several segments with direct access.

The rest of the paper is organized as follows. The proposed MISS is introduced in Section 2. Multi-scale representation of the map is first discussed in Section 2.1, and the compression method briefly recalled in Section 2.2. The decomposition of the image into binary layers is studied in Section 2.3. The image tiling into blocks for supporting efficient panning in studied in Section 2.4. The file structure and the proposed system architecture are summarized in Section 2.5. Experimental results are given in Section 3 to demonstrate the compression performance and the decoding efficiency of the system in real-time environment. Conclusions are drawn in Section 4.