Bit allocation for lossy image set compression

Abstract

Large sets of similar images are produced in many applications. To store these images more efficiently, redundancy among similar images need to be exploited. A number of methods have been proposed to reduce such inter-image redundancy in lossy image set compression. These methods encode each image either using a conventional image compression algorithm, or predicts the image from a similar image already encoded and encode the prediction residual. Although these methods differ in the way they determine the prediction structure in the image set, they do not consider the effect of bit allocation on the overall quality of the reconstructed images. In this paper, we show that Lagrangian optimization can be used to determine bit allocation for each encoded image in order to improve the overall quality of the reconstructed image set. Furthermore, a model approximating rate-distortion curves of the residual images can be used to reduce the encoding time significantly.