Related Experiment Videos
Universal image compression using multiscale recurrent patterns with adaptive probability model
E B de Lima Filho1, E B da Silva, M B de Carvalho
1Centro de Ciencia, Technologia e Inovacao, Brazil. eddie@ctpim.org.br
Summary
This study enhances the multidimensional multiscale parser (MMP) algorithm for data compression. New methods improve compression of smooth images while maintaining performance on complex data.
Area of Science:
- Computer Science
- Signal Processing
- Data Compression
Background:
- The multidimensional multiscale parser (MMP) is a universal lossy compression method.
- MMP utilizes approximate multiscale pattern matching and a self-updating dictionary.
- Existing MMP has shown success in compressing images, video, and ECG signals.
Purpose of the Study:
- To improve the MMP algorithm by incorporating source-specific knowledge.
- To enhance compression efficiency for specific data types, such as smooth images.
- To refine MMP's universal character by boosting performance on smooth data.
Main Methods:
- Developed context models based on smoothness constraints for block probabilities.
- Integrated knowledge of original block scales into the dictionary updating process.
- Applied data-specific extensions to the base MMP algorithm.
Main Results:
- Achieved significant compression improvements for smooth images compared to the original MMP.
- Maintained state-of-the-art compression performance for complex, non-smooth images.
- Demonstrated enhanced effectiveness of MMP with smoothness assumptions.
Conclusions:
- The enhanced MMP algorithm offers superior compression for smooth images.
- The developed methods improve MMP's adaptability and performance across diverse data types.
- Incorporating source knowledge strengthens the universal applicability of MMP.