Partial-Image CNN-RNN Processing for Low-Power Object Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing methods are inefficient in low-power environments due to high memory requirements and latency, especially when performing tasks like object detection, as they typically process entire frames rather than partial regions.

Innovation Solution

A method and apparatus that processes images using a convolutional neural network (CNN) and recurrent neural network (RNN) to analyze partial image units, reducing memory usage and latency by processing each partial region sequentially, allowing for efficient object detection even in low-power conditions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If entire image frames are processed using conventional neural networks, then object detection accuracy is maintained, but memory consumption and processing latency increase significantly

Engineering Contradiction:
Improveobject detection accuracyVSAvoidmemory consumption
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent divides the input image frame into multiple partial image units (e.g., 4 quadrants) and processes each unit separately through the neural network. This segmentation allows the system to maintain detection accuracy by analyzing all regions while reducing memory consumption by processing only necessary portions rather than loading the entire image into memory simultaneously.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system processes partial image units rather than the complete image frame. By applying the neural network to only the necessary partial regions (especially when objects are detected in specific quadrants), the system achieves sufficient detection accuracy without the excessive memory consumption that would result from processing the entire image.

Inventive Principle:
Principle #16Partial or excessive action

2Quantity of substance

If entire image frames are processed sequentially, then memory usage is reduced, but processing time and latency increase

Engineering Contradiction:
Improvememory usageVSAvoidprocessing latency
Core Design Contradiction:
Quantity of substanceVSLoss of time

Solution Approach 1:

By segmenting the image into partial units that can be processed in parallel, the system reduces the sequential processing time. Multiple partial image units can be handled simultaneously through different neural network instances or threads, significantly reducing overall latency while maintaining low memory usage through selective processing.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system performs preliminary processing by dividing the image into quadrants and pre-determining which regions need full neural network analysis. This preliminary segmentation allows for more efficient processing by avoiding unnecessary sequential analysis of empty regions while maintaining fast response times.

Inventive Principle:
Principle #10Preliminary action

3Quantity of substance

If partial image units are processed using neural networks, then memory consumption and latency are reduced, but computational complexity increases

Engineering Contradiction:
Improvememory consumptionVSAvoidcomputational complexity
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent segments the computational task into multiple smaller neural network processing units, each handling a specific quadrant. This segmentation reduces the complexity of any single processing unit while the overall system handles the complete image. The modular approach makes the system more manageable and easier to implement with limited computational resources.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3961494B1Method and apparatus for processing image based on partial images
Publication Date: 2025.10.29 SAMSUNG ELECTRONICS CO LTD
  • EP3961494B1 patent drawingFigure 1
  • EP3961494B1 patent drawingFigure 2
  • EP3961494B1 patent drawingFigure 3

AI summary

A method and apparatus for processing an image based on partial images. The method includes extracting a feature of a current partial processing region of an input image frame by inputting pixel data of the current partial processing region into a convolutional neural network (CNN), updating a hidden state of a recurrent neural network (RNN) for a context between the current partial processing region and at least one previous partial processing region by inputting the extracted feature into the RNN, and generating an image processing result for the input image frame based on the updated hidden state.