Image Processing Apparatus Reducing Frame Memory Access Traffic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing frame-recursive noise reduction methods require excessive access to frame memory, leading to bus bandwidth overload, as they need to access the frame memory twice for each processing target region to generate motion-compensated images using both global and local vectors.
Innovation Solution
An image-processing apparatus that calculates discrete motion vectors and confidences for a standard image and a reference image, selects between global and local vectors based on confidence and differences, and generates motion-compensated images using the selected vector, allowing for weighted addition and noise reduction while accessing the frame memory only once per processing target region.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If both global vector and local vector are used to generate motion-compensated images, then noise reduction quality is improved, but frame memory access traffic increases causing bus bandwidth overload
Solution Approach 1:
The patent applies partial action by selectively using either global vector or local vector for motion compensation based on confidence levels, rather than always using both vectors. The motion vector selecting portion chooses the appropriate vector type (global or local) depending on the confidence of the local vector, thereby reducing unnecessary frame memory accesses while maintaining noise reduction quality where sufficient confidence exists.
2Measurement precision
If discrete motion vectors are calculated for multiple measurement regions, then motion compensation accuracy is improved, but calculation complexity and processing time increase
Solution Approach 1:
The patent segments the image into multiple discrete measurement regions and calculates motion vectors for each region independently. This segmentation allows parallel processing of motion vector calculations across different regions, improving motion compensation accuracy through localized motion analysis while enabling efficient computation through distributed processing.
Solution Approach 2:
The patent performs preliminary calculation of discrete motion vectors for multiple measurement regions before generating the final motion-compensated image. By pre-calculating these vectors and their confidences, the system prepares all necessary motion information in advance, reducing real-time processing complexity during the actual noise reduction operation.
3Productivity
If local vector is used for motion compensation, then processing speed is improved, but accuracy decreases in regions with low confidence
Solution Approach 1:
The patent implements dynamic vector selection by adjusting the choice between local and global vectors based on the confidence level of the local vector. When local vector confidence is high, the system uses local vectors for faster processing; when confidence is low, it switches to global vectors for higher accuracy. This dynamic adaptation allows the system to optimize the balance between processing speed and accuracy according to local image characteristics.
Data Source
AI summary
An image-processor includes: a discrete-motion-vector calculator calculating motion vectors of individual measurement regions and confidences thereof based on a standard image and a reference image or a combined image generated based on the reference image; a global-vector calculator separately calculating, based on the motion vectors and the confidences thereof, a global vector of the standard image, a confidence thereof, a local vector of a processing target region in the standard image, and a confidence of the local vector; a local-vector interpolator; a local-vector-confidence interpolator; a motion-vector selector selecting the local vector or the global vector based on at least one of the local vector, the confidence thereof, the global vector, and the confidence thereof; a motion-compensated-image generator generating, based on the selected vector, a motion-compensated image of the reference image or the combined image; and a blender weighted adding the processing target region and the motion-compensated image.