Intra prediction based on Markov process modeling of images

Area of Science:

• Digital image processing
• Video compression algorithms
• Computational photography

Background:

• Current video coding standards use angular intraprediction, which is computationally efficient but overlooks pixel correlations.
• General linear prediction leverages more pixel correlations but suffers from high computational complexity due to numerous weights.

Purpose of the Study:

• To propose a computationally efficient intraprediction method for video coding.
• To leverage pixel correlations more effectively than standard angular prediction.
• To reduce the complexity associated with general linear prediction methods.

Main Methods:

• Modeling image pixels using a Markov process for intraprediction.
• Developing a computationally efficient recursive prediction algorithm based on the Markov model.
• Comparing the proposed method against general linear prediction and standard intraprediction.

Main Results:

• The Markov process model accounts for ignored correlations in standard intraprediction.
• The proposed method achieves computationally efficient recursive prediction.
• It uses fewer parameters than general linear prediction, significantly reducing memory and computation requirements.
• Similar coding gains are achieved compared to general linear prediction with pre-computed parameters.

Conclusions:

• The Markov process modeling approach offers a computationally efficient and effective alternative for intraprediction in video coding.
• It balances the need for utilizing pixel correlations with computational feasibility.
• This method presents a significant improvement over existing intraprediction techniques in terms of efficiency and performance.