IBR has been widely used in computer games and virtual navigation in recent years. For these applications, we need to render the pictures of novel views in real time. Over the last few years, many researchers have made contributions to this aspect. For example, LDI rendering algorithm possesses advantages in rendering speed and quality. These algorithms directly use perspective projection during rendering so that the rendering speed is relatively slow.

In volume rendering, shear-warp rendering is a representative algorithm and widely used by many applications in practice. For regular 3D grid data, it first generates an intermediate image by a shear transformation, and then it warps the intermediate image to get the final image by using a 2-D warp transformation. Besides software implementation, the warping procedure can also be done by using texture hardware.

By combining the advantages both layer-based algorithm and shear-warp algorithm, we present a new IBR algorithm, which can achieve interactive rendering speed and high rendering quality. For reference images with depth information, we classify all the pixels into six types according to their sight line directions; then we arrange all the pixels in each type into parallel layers by utilizing ISODATA cluster algorithm and weak perspective principle, and compress data of each layer type by RLE encoding. During rendering, for any novel viewpoint, we chooses the corresponding layer types to render an intermediate image based on shear transformation, and further generates the final image from the intermediate one by using warp transformation. Our work introduces shear-warp volume rendering algorithm into IBR for the first time, which is different from the traditional IBR algorithms that directly use perspective projection transform during rendering. We compare our algorithm with the LDI rendering algorithm. The experiments show that our algorithm can not only generate high quality images but also achieve interactive frame rates.