Saliency Map Evaluation via Distribution Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image-processing technologies fail to accurately identify and evaluate salient regions in images based on saliency maps, as they rely solely on feature quantity scores or local maximum points, lacking a comprehensive approach to differentiate between salient and non-salient regions.
Innovation Solution
An image-processing apparatus and method that calculates saliency maps from input images using luminance, color, and texture features, identifies salient regions, and evaluates their saliency by comparing distribution scores between salient and non-salient regions, incorporating weighted averages, standard deviations, and histogram comparisons to determine region saliency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing image-processing technologies use feature quantity scores or local maximum points to identify salient regions, then the processing is simple and fast, but the accuracy of salient region identification is insufficient
Solution Approach 1:
The patent segments the saliency evaluation process into multiple independent components: feature quantity extraction (luminance, color, texture), saliency map generation for each feature, salient region identification, and distribution-based score calculation. This segmentation allows each component to be optimized independently while maintaining overall accuracy.
Solution Approach 2:
The patent changes the evaluation parameter from simple feature scores or local maximum points to a comprehensive distribution-based score that compares statistical properties (mean, standard deviation) of saliency values between salient and non-salient regions. This parameter change significantly improves identification accuracy.
2Reliability
If existing technologies rely on single feature quantity scores, then the processing is computationally efficient, but the ability to differentiate between salient and non-salient regions is limited
Solution Approach 1:
The patent merges multiple feature quantities (luminance, color, texture) and their corresponding saliency maps into a unified evaluation framework. The salient region score integrates information from all features by comparing the distribution of saliency values across features, enhancing the differentiation capability between salient and non-salient regions.
Solution Approach 2:
The patent creates a composite saliency evaluation metric that combines multiple feature types and statistical measures (mean, standard deviation). This composite approach is analogous to using composite materials, where the combination of different feature types produces a more robust and reliable salient region identification than any single feature alone.
Data Source
AI summary
An image-processing apparatus including: an image processor including circuitry or a hardware processor that operates under control of a stored program, the image processor being configured to execute processes including: a saliency-map calculating process that calculates saliency maps on a basis of at least one type of feature quantity obtained from an input image; a salient-region-identifying process that identifies a salient region by using the saliency maps; a salient-region-score-calculating process that calculates a score of the salient region by comparing a distribution of values of the saliency map in the salient region and a distribution of values of the saliency map in a region other than the salient region; and a saliency-evaluating process that evaluates the saliency of the salient region on a basis of the score.