Image Processing Apparatus Constrained Intra-Prediction Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques face challenges in suppressing deterioration in image quality due to the accumulation of lossy compression effects in the temporal direction of image decoding, particularly when no lossy compression of reference data is performed, leading to increased image quality degradation.
Innovation Solution
An image processing apparatus and method that utilize a predicated image generator to generate images using inter-prediction and constrained intra-prediction, where the inter-prediction references encoded and decoded images from a frame memory, and the constrained intra-prediction does not use neighboring pixels subjected to inter-prediction, allowing for adaptive selection of reference frames and control of prediction methods based on frame differences and coding difficulty levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If lossy compression is applied to reference data in frame memory, then data storage capacity and data bandwidth requirements are reduced, but image quality deterioration accumulates in the temporal direction
Solution Approach 1:
The reference frame is divided into two distinct types: long-term reference frames stored in frame memory and short-term reference frames stored in buffer memory. This segmentation allows selective application of compression strategies - lossy compression for long-term frames to reduce memory requirements, and lossless preservation for short-term frames to maintain prediction accuracy, thereby resolving the contradiction between storage efficiency and image quality
Solution Approach 2:
A dual-memory system acts as an intermediary between the frame memory and buffer memory, enabling controlled data flow and selection between different reference frame types. This intermediary mechanism allows the system to balance compression benefits against quality requirements by selectively accessing appropriate memory sources based on prediction needs
2Manufacturing precision
If lossless compression is applied to reference data, then image quality deterioration is suppressed, but data storage capacity and data bandwidth requirements increase
Solution Approach 1:
Different quality levels are applied to different parts of the reference frame system: lossless quality for short-term reference frames in buffer memory and lossy quality for long-term reference frames in frame memory. This local differentiation optimizes overall system performance by applying appropriate compression strategies to specific memory regions based on their functional requirements
3Manufacturing precision
If no lossy compression is applied to reference data, then image quality is maintained, but the complexity of access methods to reference memory increases
Solution Approach 1:
The system dynamically selects between long-term and short-term reference frames based on prediction mode and timing conditions. The prediction processing unit adaptively determines which reference frame type to use, allowing flexible access patterns that balance quality maintenance with manageable complexity through intelligent control rather than rigid fixed-access methods
4Productivity
If conventional inter-prediction is used without constrained intra-prediction, then decoding complexity is reduced, but image quality deterioration accumulates more rapidly
Solution Approach 1:
The patent merges inter-prediction and intra-prediction into a unified constrained intra-prediction mode that operates on previously decoded pixel values. This combination allows the system to leverage both temporal (inter-frame) and spatial (intra-frame) prediction benefits while maintaining decoding efficiency, thereby improving image quality without proportionally increasing decoding complexity
Data Source
AI summary
The present disclosure relates to image processing apparatus and method intended to be capable of suppressing deterioration in image quality. [Solution] A predicated image is generated using inter-prediction and constrained intra-prediction, the inter-prediction using as a reference image an image of a past frame processed before a current frame, the constrained intra-prediction using no neighboring pixels subjected to inter-prediction, the image of the past frame being encoded, stored in a frame memory, read from the frame memory, and then decoded. The present disclosure is applicable to, in one example, an image processing apparatus, an image coding apparatus, or the like.