Multi-core DSP Loop Filtering via Macro Block Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High definition image processing requires significant processing resources, and current single chip core processors are insufficient for independent high definition image encoding, leading to inefficiencies in multi-core image encoding devices, particularly in the loop filtering process of the H.264 protocol, where one DSP chip must wait for another to complete filtering before proceeding.
Innovation Solution
Implementing a method where multiple DSP chips in a multi-core image encoding device perform loop filtering concurrently by skipping the highest horizontal boundary of the first row of macro blocks, allowing each chip to start filtering without waiting for the previous chip to complete, thereby improving resource utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If loop filtering is performed strictly according to the grating scan order of macro blocks as defined in H.264, then filtering correctness is ensured, but DSP resources cannot be utilized efficiently because one DSP chip must wait for the previous chip to complete filtering
Solution Approach 1:
The patent divides the loop filtering task into two independent segments: one DSP chip performs filtering on even-numbered macro blocks while another DSP chip performs filtering on odd-numbered macro blocks. This segmentation allows both chips to work simultaneously without waiting for each other, resolving the contradiction between maintaining filtering correctness and improving DSP resource utilization efficiency.
2Ease of operation
If an image frame is horizontally divided into upper half and lower half for load balance, then visual effect is improved, but filtering must be performed in series between DSPs causing waste of DSP resources
Solution Approach 1:
Instead of dividing the image frame into upper and lower halves, the patent segments the filtering task by macro block numbering (even vs. odd). This allows DSP chips to process different segments simultaneously without series execution, eliminating wait time while maintaining load balance and visual quality.
Solution Approach 2:
The patent inverts the traditional sequential filtering approach by allowing out-of-order processing of even and odd macro blocks simultaneously. Rather than having DSP1 wait for DSP0 to complete, both chips work in parallel on their respective segments, then results are combined to produce the final filtered image.
Data Source
AI summary
An embodiment of the present invention discloses a method for performing image filtering by a multi-core image encoding device, comprising: a first digital signal processing (DSP) chip performing loop filtering on the image encoded by itself; and the latter digital signal processing (DSP) chip in any two adjacent DSP chips performing loop filtering on all horizontal boundaries except the highest horizontal boundary and all vertical boundaries of the first row of macro blocks in the image encoded by itself. An embodiment of the present invention further discloses a multi-core image encoding device. Using the method for performing image filtering of the present invention, utilization of DSP resources in the multi-core image encoding device can be improved effectively.