Iterative Near-Lossless Image Compression With Entropy Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image compression techniques face limitations in achieving significant compression gains while maintaining acceptable quality, particularly in near-lossless methods that require controlled data loss within bounded reconstruction error.
Innovation Solution
An iterative method for compressing image data that updates entropy models after each compression iteration, allowing decisions on data loss based on the revised entropy models to ensure that the loss remains within acceptable bounds, thereby balancing compression gains and quality degradation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If lossless compression techniques are used to ensure exact reconstruction of original data, then data quality is preserved, but compression gains are limited
Solution Approach 1:
The patent changes the parameter of data loss tolerance from zero (lossless) to a controlled bounded amount (near-lossless). By introducing a loss bound parameter and using entropy models to guide selective data loss decisions, the system achieves better compression ratios while maintaining quality within acceptable thresholds through iterative compression processes.
Solution Approach 2:
The patent implements dynamic adaptation through iterative compression cycles where entropy models are updated based on previous compression results. The system dynamically adjusts compression decisions in each iteration based on the current entropy model, allowing it to optimize the balance between compression gain and quality preservation adaptively rather than using fixed parameters.
2Productivity
If lossy compression techniques are used to achieve greater compression gains, then compression efficiency is improved, but data quality degradation increases
Solution Approach 1:
The patent implements feedback mechanisms through entropy models that are updated after each compression iteration based on the actual compression results and data loss patterns. This feedback loop allows the system to learn from previous compression decisions and adjust subsequent decisions to maintain quality within bounded loss thresholds while maximizing compression gains.
Solution Approach 2:
The patent performs preliminary actions by building and updating entropy models before final compression decisions are made. These entropy models predict the impact of potential data loss on compression efficiency, allowing the system to pre-determine optimal compression strategies that will achieve desired compression gains while staying within quality bounds.
3Productivity
If near-lossless compression is used to allow some data loss for better compression, then compression gains increase, but controlling reconstruction error becomes more complex
Solution Approach 1:
The patent introduces entropy models as intermediary components that mediate between the compression process and quality control requirements. These entropy models serve as a bridge by predicting compression outcomes and guiding loss decisions, simplifying the overall error control mechanism while enabling better compression gains through automated, model-based decision-making.
Data Source
AI summary
Systems and methods for iterative near lossless image compression are provided. An exemplary computer-implemented method of compressing image data can include performing a plurality of compression iterations. Each compression iteration can include at least one decision regarding a loss of image data. The method can also include updating an entropy model following each compression iteration. The entropy model can describe an entropy associated with selected of a plurality of data blocks. The at least one decision regarding the loss of image data can be decided, for each compression iteration, based on the entropy model as updated following the previous compression iteration. Further, a total loss of image data from each data block can remain within an acceptable loss bound associated with the pixel described by such data block. An exemplary system can include a loss determination module, an entropy modeling module, a compression module, and an entropy coding module.