Efficient Prediction Technique of Zero Quantized DCT Coefficients for Fast Image Encoding
Résumé
In this paper we present a new method to predict Zero Quantized Discrete Cosine Transform (ZQDCT) coefficients for fast image encoding. Traditional methods use the Sum of Absolute Difference (SAD) of residual pixels to predict all-zero or partial-zero blocks. The proposed method presents two main differences. First, bounds of DCT coefficients are derived by considering the intermediate signals of the Loeffler DCT algorithm instead of calculating the SAD of residual pixels. Second, the derived sufficient conditions allow to predict individual zero quantized coefficients to increase the number of predicted ZQDCT coefficients. Simulations corroborate the validity of the analytical model and show that ZQDCT detected coefficients are averagely about 50% times more than in the existing method