Image Processing Apparatus Memory Bandwidth Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems that perform cyclic noise reduction (NR) and rotation processing sequentially consume large amounts of memory bandwidth and area, leading to system performance deterioration and increased costs, especially in high-resolution video processing.
Innovation Solution
The image processing apparatus includes dedicated units for cyclic NR processing and inversion processing, which allow for efficient memory usage by omitting storage of frame data after NR processing and before inversion, and by controlling the read and write addresses to match the phases of frames, reducing memory bandwidth and area requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cyclic NR processing and rotation processing are performed sequentially using memory for frame data storage, then image processing functionality is achieved, but memory bandwidth and area are excessively consumed
Solution Approach 1:
The patent combines cyclic NR processing and rotation processing into a unified processing pipeline. The geometric conversion unit performs rotation on frame data, and the cyclic NR unit immediately processes the rotated data without requiring separate memory storage steps. This merging eliminates intermediate memory writes and reads, reducing memory bandwidth consumption while maintaining both processing functions.
Solution Approach 2:
The patent introduces an intermediary processing structure where the geometric conversion unit acts as a bridge between memory and the cyclic NR unit. Instead of writing rotated data back to memory, the system creates a direct data flow path where the intermediary unit (geometric conversion) feeds directly into the next processing unit (cyclic NR), eliminating the need for intermediate memory storage and reducing memory area requirements.
2Adaptability or versatility
If frame data is written to and read from memory for each processing step, then processing flexibility is maintained, but system performance deteriorates due to excessive memory accesses
Solution Approach 1:
The patent segments the image processing system into specialized functional units: a geometric conversion unit for rotation operations and a cyclic NR unit for noise reduction. Each unit is optimized for its specific function and processes data in a dedicated pipeline. This segmentation allows each unit to operate independently and efficiently without requiring repeated memory accesses, thereby improving system performance while maintaining processing flexibility through modular design.
Solution Approach 2:
The geometric conversion unit performs rotation processing as a preliminary action before the cyclic NR unit processes the data. By pre-rotating the frame data in the geometric conversion unit and directly passing it to the cyclic NR unit, the system eliminates the need to store intermediate rotated data in memory. This preliminary action approach maintains processing flexibility while significantly reducing memory access overhead and improving overall system performance.
Data Source
AI summary
An image processing apparatus including an input unit configured to input image data of a target frame, a first processing unit configured to perform a first processing for output image data of a reference frame related to the target frame, which is stored in the memory, to change positions of pixels in the output image data, an image processing unit configured to perform image processing for the image data of the target frame input by the input unit based on the output image data of the reference frame processed by the first processing unit, and a second processing unit configured to perform a second processing, the second processing being according to the first processing, for the image data of the target frame for which image processing has been performed by the image processing unit and to store the processed image data in the memory as output image data of the target frame.