Image Processing Apparatus Bandwidth Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing systems consume excessive bandwidth on memory buses due to the need for multiple burst transmissions to read adjacent pixel data for filter execution, especially when processing large image blocks, leading to inefficient filter processing.
Innovation Solution
An image processing apparatus that includes a memory, decoding circuit, filter execution circuit, and memory controllers to perform filter execution block by block, utilizing a line buffer and boundary buffer to efficiently read and write data, reducing the number of burst transmissions required and minimizing bandwidth consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If image data of adjacent pixels is read using burst transmission from DRAM for filter execution, then filter processing can be performed, but the number of burst transmissions increases and bandwidth consumption increases
Solution Approach 1:
The patent divides the image processing into blocks and further segments the data retrieval operations. Instead of reading all adjacent pixel data in single large bursts, the system segments the reading into targeted operations that only fetch necessary data portions (current block plus adjacent pixels) in smaller, more efficient bursts. The line buffer and boundary buffer further segment the storage to optimize access patterns.
Solution Approach 2:
The patent performs preliminary actions by pre-reading and storing adjacent pixel data into the line buffer before filter execution begins. The boundary buffer pre-stores boundary pixel data that will be needed for subsequent blocks. This preliminary data preparation reduces the need for repeated burst transmissions during actual filter processing, as the data is already available in faster memory.
2Reliability
If all image data is decoded and stored before filter execution, then complete image data is available for processing, but memory bandwidth consumption increases due to multiple burst transmissions
Solution Approach 1:
The patent extracts and processes only the necessary portions of image data (current block plus adjacent pixels) rather than requiring all image data to be decoded and stored. The line buffer and boundary buffer extract and hold only the specific data portions needed for filter execution, reducing memory bandwidth consumption while ensuring data availability for processing.
Solution Approach 2:
The patent introduces intermediary buffers (line buffer and boundary buffer) between the DRAM and the filter execution unit. These buffers act as mediators that hold the necessary data locally, reducing the need for continuous DRAM access and multiple burst transmissions. The intermediary buffers ensure data availability for processing while minimizing bandwidth consumption by reducing memory access frequency.
3Manufacturing precision
If the filter execution requires image data of pixels adjacent to the block, then accurate filter results are obtained, but the number of burst transmissions required increases
Solution Approach 1:
The patent performs preliminary action by pre-reading and storing adjacent pixel data into the line buffer and boundary buffer before filter execution. This ensures that when filter execution requires accurate adjacent pixel data, the data is already available in the buffers, eliminating the need for additional burst transmissions during processing and maintaining both accuracy and processing speed.
Solution Approach 2:
The patent uses intermediary buffers (line buffer and boundary buffer) to store adjacent pixel data close to the filter execution unit. These buffers act as intermediaries that provide fast access to the precise adjacent pixel data needed for accurate filter execution, reducing the reliance on slow DRAM burst transmissions and thereby improving processing speed while maintaining filter accuracy.
Data Source
AI summary
A data input circuit stores decoded image data into the line buffer. A transmission circuit transmits image data of a rear end part in the primary scanning direction of each line in the line buffer to a boundary buffer. In the secondary scanning direction, a second block is adjacent to a first block, and a fourth block is adjacent to a third block. The third block is adjacent to the first block in the primary scanning direction. The filter execution circuit continuously performs filter execution of: (a) the rear end part in the primary scanning direction of (a1) the rear end part in the secondary scanning direction of the first block and (a2) the second block, and (b) the rear end part in the secondary scanning direction of the third block and (c) the fourth block, using the image data stored in the boundary buffer and the line buffer.