Bayesian Salient Region Detection in Image Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing salient region detection methods struggle to accurately identify salient regions from single images without prior information about the object and background regions, and they require multiple images for processing, leading to high computational load and lower precision.

Innovation Solution

An image processing device employing Bayesian theory to calculate posterior probabilities of salient regions by determining prior probabilities for each pixel, using superpixel images and convex hull regions, effectively utilizing Bayes' theorem for precise salient region detection from a single image.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple input images are used to extract salient region images, then the accuracy of salient region detection is improved, but the computational load and processing time increase significantly

Engineering Contradiction:
Improvesalient region detection accuracyVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent pre-calculates and stores prior probability images during an initialization phase using multiple training images. This preliminary action allows the system to perform rapid salient region extraction on single test images without repeatedly processing multiple images, thus improving processing speed while maintaining detection accuracy

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent divides the image processing into separate modules: prior probability calculation, likelihood calculation, and posterior probability computation. By segmenting the processing steps and pre-computing the prior probability component, the system reduces the computational burden during actual salient region extraction, resolving the contradiction between accuracy and processing speed

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If prior probability images are updated using salient region images from previous time points, then the detection accuracy is improved, but the device complexity and processing requirements increase

Engineering Contradiction:
Improvesalient region detection accuracyVSAvoidprocessing system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent uses prior probability images as a simplified representation (copy) of statistical information derived from multiple images. Instead of processing actual image data repeatedly, the system works with pre-computed probability maps that capture essential patterns, reducing device complexity while maintaining detection precision

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent transforms image data into probability parameter representations. By changing the parameter space from pixel intensities to probability distributions, the system simplifies the processing requirements and device complexity while improving the robustness and accuracy of salient region detection

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If a learning database is built for salient region detection, then the detection performance is improved, but the difficulty of data preparation and system setup increases

Engineering Contradiction:
Improvesalient region detection performanceVSAvoidsystem setup ease
Core Design Contradiction:
Measurement precisionVSEase of manufacture

Solution Approach 1:

The patent enables the system to automatically compute prior probability images from provided training images without requiring manual annotation or complex database construction. The system self-learns statistical patterns from unannotated images, making setup easier while maintaining detection performance

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent uses simple, unannotated training images instead of expensive, manually labeled databases. By using readily available images that don't require complex preparation or long-term maintenance, the system achieves good performance with easier setup and lower resource requirements

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

Data Source

PatentEP2797052B1Detecting a saliency region in an image
Publication Date: 2019.10.16 OMRON CORP
  • EP2797052B1 patent drawingFigure 1
  • EP2797052B1 patent drawingFigure 2~3
  • EP2797052B1 patent drawingFigure 4

AI summary

This invention realizes an image processing device or the like capable of detecting a salient region from one image with precision by effectively using Bayes' theorem even when prior information regarding the salient region is not given. A prior probability calculating unit calculates prior probability of a salient region so as to increase prior probability of each of pixels constructing a superpixel whose color distance from the periphery is large, a likelihood calculating unit calculates likelihood of the salient region, and a salient region detecting unit calculates posterior probability of the salient region by using the prior probability and the likelihood.