Layered image resizing in compression domain

As an efficient unitary transform, the discrete cosine transform (DCT) has been widely adopted in compression standards. Most compressed images and videos are stored in the DCT format, and from time to time, they need to be resized for various transmission channels and consumer terminals. In this paper, we investigate existing resizing schemes, focusing on the difference of short and long basis vector truncation. A layered resizing scheme is then proposed based on the above analysis, where compressed images are divided into low- and high-frequency layers. The DCT vectors of these two layers are truncated with different word lengths, and then form the elementary layer (EL) and the enhancement layer (EH) of the downsampled image. The EL and EH can be transmitted together or separately according to the bandwidth available. An upsampling scheme is also provided in this paper to recover visual details. Experimental results show improvements of the proposed approach over existing resizing schemes, which can be explained since its frequency response is closer to the ideal downsample filter. The new approach can cater for user terminals’ display limitation and channel bandwidth constraints, with additional scalability.