Image Encoding Apparatus Tile-Based Quantization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge is to record RAW image data at a compact size while maintaining image quality without loss, as existing methods require a large number of quantization parameters due to the high data amount generated from two-dimensional discrete wavelet transforms of color channels in image sensors.
Innovation Solution
The solution involves dividing RAW image data into tiles, generating channels with different components, frequency-transforming the data, and applying common quantization parameters to segments of sub-band data to reduce the data amount of quantization parameters, thereby enabling compact recording while maintaining image quality through flexible quantization control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If two-dimensional discrete wavelet transform is applied to each color channel to generate multiple sub-bands for lossy compression, then compression efficiency is improved, but the number of quantization parameters required becomes extremely high
Solution Approach 1:
The patent merges the quantization parameters across multiple color channels and sub-bands by determining quantization parameters based on the maximum value among all channels and sub-bands in a tile. This consolidation reduces the total number of quantization parameters from potentially hundreds to a manageable set, while still maintaining the benefits of multi-channel compression through the wavelet transform.
Solution Approach 2:
The patent changes the approach to quantization parameter determination from individual channel/sub-band parameters to a unified parameter set based on maximum values across all channels and sub-bands. This parameter transformation enables lossy compression to proceed with significantly reduced parameter complexity, resolving the contradiction between compression efficiency and parameter quantity.
2Measurement precision
If quantization parameters are determined individually for each color channel and sub-band, then quantization control precision is improved, but the data amount of quantization parameters becomes extremely high
Solution Approach 1:
The patent creates a universal quantization parameter determination method that applies to all color channels and sub-bands through a single tile-based approach. By using the maximum value across all channels and sub-bands in each tile as the basis for quantization parameters, the system achieves multi-functional quantization control without requiring separate parameters for each channel and sub-band, thus reducing the overall data amount while maintaining precision.
Solution Approach 2:
The patent segments the image into tiles, where each tile independently determines its quantization parameters based on the maximum value of all color channels and sub-bands within that tile. This segmentation allows for localized quantization control while sharing the burden of parameter determination across multiple channels and sub-bands, reducing the total number of parameters needed compared to individual parameter determination for each channel and sub-band.
3Productivity
If RAW data is divided into multiple tiles for encoding, then processing efficiency is improved, but image quality may drop at tile boundaries
Solution Approach 1:
The patent applies preliminary action by extending the generating unit's operation to include not only the current tile but also adjacent tiles within a predetermined distance from the boundary. This preliminary data gathering from adjacent tiles allows the system to prepare and smooth transition data at boundaries before encoding, preventing quality drops while maintaining the processing efficiency benefits of tile division.
Solution Approach 2:
The patent introduces an intermediary approach by considering adjacent tiles as mediators in the data generation process. The generating unit uses data from adjacent tiles (within a predetermined distance) as intermediary information to supplement the current tile data, ensuring smooth transitions and maintaining image quality at boundaries while still encoding each tile independently for processing efficiency.
Data Source
AI summary
An encoding apparatus comprises a dividing unit that divides RAW image data into a plurality of tiles, a generating unit that generates planes of a plurality of channels having mutually-different components for each of the tiles, a transforming unit that frequency-transforms the plane of each channel and generates sub-band data of a plurality of resolution levels, a controller that divides each of the plurality of sub-band data into the same number of segments to divide the sub-band data into a plurality of segments corresponding to the same regions, and determines first quantization parameters common for the plurality of sub-band data in each segment, a quantizing unit that, based on the determined first quantization parameters, quantizes each sub-band data, and an encoder that encodes a quantization result on a sub-band-by-sub-band basis.