Remote visualization of an arbitrary 2-D planar "cut" from a large volumetric dataset with random access has both gained importance and posed significant challenges over the past few years in industrial and medical applications. In this paper, a prediction model is presented that relates transmission efficiency to voxel coverage statistics for a fast random 2-D image retrieval system. This model can be used for parameter selection and also provides insights that lead us to propose a new 3D rectangular tiling scheme, which achieves an additional 10% -30% reduction in average transmission rate as compared to our previously proposed technique, e.g., a nearly 30% / 45% reduction in the average transmission rate at the cost of a factor of ten / fifteen in storage overhead compared to traditional cubic tiling. Furthermore, this approach leads to improved random access, with less storage and run-time memory required at the client.I. INTRODUCTION Researchers in many fields (e.g., biomedical imaging, earth sciences, computational fluid dynamics, etc.) need to manipulate and visualize very large datasets. In these kinds of applications, a client-server approach has been often used in order for a personal computer equipped with limited memory to be able to manipulate, visualize, and render the complete dataset. With a client-server approach (e.g., [1]-[6]), the server provides only the data needed for the specific visualization task required at the client end. We focus on situations where lower dimensional portions of a dataset need to be accessed. In this paper, arbitrary oblique planes of a 3D volume may need to be extracted and rendered, as is required in some medical imaging applications. An example of oblique plane intersection with the 3D volume is shown in Fig. 1. In many techniques proposed for volumetric image coding [2]-[6], including approaches such as JP3D [7]-[9], some form of random access is provided via a non-overlapped, independently encoded, cuboid tiling. These approaches can be inefficient in the scenario we focus on, because the only useful voxels 1 for each retrieved cubic tile are those near the intersection between the cube and the desired 2D plane. In our previous work [10], we showed that it is more efficient to use overlapping rotated tiles to represent the dataset, which leads to an increase in the average number of useful voxels per tile so that the total number of tiles to be retrieved is smaller (and hence a lower transmission bitrate is achieved). In this approach, we trade-off increased storage at the server's side for lower bandwidth during interactive access to the data-set. Our proposed system is displayed in Fig. 2 and consists of 5 components: 1) 3D rectangular tiling scheme, 2) mapping algorithm, 3) tile searching algorithm, 4) 3D compression, and 5) random oblique plane reconstruction and display. We have proposed a 3D rectangular tiling scheme and discussed components (2)