A new adaptive coefficient scanning based on local and global prediction

Abstract

In this paper, a new adaptive coefficient scanning scheme, which is called local- and global-prediction-based adaptive scanning (LGPAS), is described to improve the coding efficiency of discrete cosine transform (DCT)-based image compression methods including JPEG and H.264/AVC intra-coding, in which zigzag scanning is used. The coding performance is limited because the zigzag scan order ignores the statistical properties of the DCT coefficients. On the other hand, we adopt not only the global information but also the local information to perform learning and adaptively generate the scanning patterns, unlike the existing methods. Furthermore, we adopt variation prediction, nonzero probability estimation, and the proposed techniques of zigzag weighting and energy weighting matrices to generate the scanning pattern. On the basis of the local and global predictions for the probability distributions of the nonzero DCT coefficients in an image, the proposed LGPAS scheme can adaptively update the scan order patterns and thus achieves a higher entropy coding gain. Simulations show that the proposed scheme significantly outperforms the conventional zigzag scanning method and other existing adaptive scanning methods.