Image Processing Device Data Reconstruction Circuit Parallel Depthwise Convolution

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing technologies face performance issues due to the need for a large number of multiplication and addition operations when processing large input image data or filter masks, leading to reduced overall performance.

Innovation Solution

An image processing device and method that utilize a data reconstruction circuit to convert image data and filter data into reconstruction data, allowing an intelligent processor to perform depthwise convolution operations in parallel, thereby accelerating image processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If image filtering is implemented by sequential multiplication and addition operations performed by a processor, then the processing approach is simple and easy to implement, but the processing speed is slow and overall performance is reduced when dealing with large input image data or filter masks

Engineering Contradiction:
Improveimage processing speedVSAvoidprocessing architecture complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent divides the image data into multiple blocks and reconstructs them into an expanded data structure. This segmentation allows the intelligent processor to handle multiple blocks simultaneously through depthwise convolution operations, transforming a sequential processing approach into a parallel one, thereby improving image processing speed without requiring a completely new processing architecture

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transforms the original image data structure by reconstructing multiple blocks into an expanded dimensionality format. This dimensional transformation enables the intelligent processor to perform depthwise convolution operations that process multiple image blocks in parallel, effectively adding a processing dimension and significantly accelerating image filtering performance

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Loss of time

If the processor performs only one multiplication or addition operation in one single period, then the operation is simple and precise, but a considerable number of periods are required to complete filtering when the size of input image data or filter mask is large

Engineering Contradiction:
Improvefiltering completion timeVSAvoidoperations per period
Core Design Contradiction:
Loss of timeVSProductivity

Solution Approach 1:

The patent performs preliminary reconstruction of image data blocks before the actual filtering operation. By pre-organizing multiple image blocks into an expanded data structure suitable for depthwise convolution, the system prepares the data in advance to enable parallel processing during the filtering stage, reducing the total time required to complete the filtering operation

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent changes the data structure parameters by reconstructing multiple image blocks into an expanded format with specific dimensional relationships. This parameter transformation allows the intelligent processor to execute depthwise convolution operations that process multiple blocks simultaneously, increasing the effective operations per period and reducing filtering completion time

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS12229850B2Image processing device and image processing method
Publication Date: 2025.02.18 SIGMASTAR TECH LTD
  • US12229850B2 patent drawing
  • US12229850B2 patent drawing
  • US12229850B2 patent drawing

AI summary

An image processing device performs a filtering operation on image data according to a set of filter data, and includes a data reconstruction circuit, a memory and an intelligent processor. The data reconstruction circuit reconstructs data of blocks in the image data to generate first reconstruction data, and reconstructs the set of filter data to generate second reconstruction data. The memory stores the first reconstruction data and the second reconstruction data. The intelligent processor executes a depthwise convolution operation according to the first reconstruction data and the second reconstruction data to generate feature map data. Two adjacent of the multiple blocks have partially same data, and quantities of columns in the second reconstruction data are associated with a channel count of the depthwise convolution operation.