Image Encoding Using Modality-Specific Probability Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image encoding methods struggle to achieve high compression ratios while maintaining image quality, especially in wireless transmission where data losses are common, and they lack adaptability to different image modalities.
Innovation Solution
The method involves segmenting images into uniform blocks, applying a frequency-based transform, quantizing coefficients, and converting them into binary code using a probability model learned from representative images, which adapts to different image modalities through pre-filters and optimized probability models for enhanced compression and robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional compression algorithms are used to reduce data size, then transmission speed improves, but compression ratio and image quality deteriorate
Solution Approach 1:
The patent applies parameter changes by using modality-specific probability models that adapt to different image types (infrared, visible, hyperspectral). By learning and storing probability models for each modality, the system optimizes the coding parameters for binary arithmetic coding, achieving higher compression ratios without quality loss. This resolves the contradiction by enabling efficient compression tailored to specific image characteristics rather than using generic algorithms.
Solution Approach 2:
The patent implements preliminary action by pre-learning probability models from representative images of each modality and storing them for later use. This pre-processing step allows the encoding system to immediately apply optimized probability models when encoding actual images, achieving high compression ratios from the start without requiring iterative optimization during the encoding process itself.
2Device complexity
If generic probability models are used for binary arithmetic coding, then device complexity is reduced, but compression ratio deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the probability modeling into modality-specific segments. Instead of using a single generic probability model for all images, the system creates and stores separate probability models for different image modalities (infrared, visible, hyperspectral). This segmentation allows each model to be optimized for its specific modality, achieving high compression ratios while keeping individual models relatively simple in structure.
Solution Approach 2:
The patent implements dynamics by making the probability model selection adaptive and dynamic. The encoding system dynamically selects the appropriate pre-learned probability model based on the modality of the input image. This dynamic adaptation allows the system to achieve optimal compression ratios for each image type without requiring complex real-time model learning, balancing simplicity and performance.
3Loss of information
If image-specific probability models are learned and stored, then compression ratio improves, but device complexity increases
Solution Approach 1:
The patent applies preliminary action by performing the complex probability model learning process in advance, before actual image encoding. Representative images for each modality are used to learn and store probability models during a pre-processing phase. This shifts the computational complexity from the encoding phase to the model preparation phase, allowing the encoding system to simply retrieve and apply pre-computed models, thus achieving high compression ratios without burdening the encoding device with complex real-time computations.
4Ease of operation
If uniform block size is used for segmentation, then processing simplicity is maintained, but adaptability to different image modalities deteriorates
Solution Approach 1:
The patent applies parameter changes by maintaining uniform block sizes for segmentation while changing the probability model parameters to adapt to different modalities. The block segmentation structure remains simple and consistent across all image types, but the probability models applied to these blocks are modality-specific. This allows the system to maintain processing simplicity through uniform segmentation while achieving modality adaptability through parameter changes in the probability models.
Data Source
Figure 1a~1B
Figure 2~6
Figure 7~8
AI summary
A method for encoding data defining an image is disclosed. The image is segmented into image blocks, each image block having a uniform block size. A frequency-based transform is applied to each of the image blocks, thereby providing transformed image data in which the image data is represented as coefficients defining a linear combination of predetermined basis functions having different spatial frequencies. The coefficients are quantised and converted into binary code. The conversion comprises application of binary arithmetic coding using a probability model. The probability model is learnt based on a sample set of representative images.