Weighted Feature Image Representation Using Saliency Maps

Solution Overview

Problem

Existing image processing methods for categorization and retrieval often disregard background information, are computationally costly, and lack robustness due to reliance on interest point detectors and foreground-based approaches.

Innovation Solution

A method that generates a weighted feature-based image representation by incorporating location relevance information, using gaze data or saliency maps to assign higher weights to features in regions of high relevance, allowing for effective use of both keypoint-based and continuous saliency maps within a unified framework.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If interest point detectors are used for feature extraction, then reliable correspondences and matching are obtained, but background regions carrying important category information are disregarded and computational cost increases

Engineering Contradiction:

Improvematching reliabilityVSAvoidcategorization accuracy

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The patent applies local quality by weighting features differently based on their spatial location and relevance. Instead of treating all features equally, the method assigns higher weights to features in regions identified as salient or relevant through saliency maps, while downweighting or ignoring features in less relevant background regions. This resolves the contradiction by maintaining the reliability of key feature matching while simultaneously capturing background information that carries category information.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent segments the image into multiple regions and assigns different processing weights to each region based on saliency maps. By dividing the image into salient and non-salient regions, the method can selectively process and weight features from different regions, ensuring that both foreground objects and background context are appropriately considered in the final representation.

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If dense sampling strategies are used instead of interest point detectors, then better performance in accuracy and speed is achieved, but computational complexity increases

Engineering Contradiction:

Improvecategorization accuracyVSAvoidcomputational complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by weighting features differently based on their spatial location and relevance. Instead of treating all features equally, the method assigns higher weights to features in regions identified as salient or relevant through saliency maps, while downweighting or ignoring features in less relevant background regions. This resolves the contradiction by maintaining the reliability of key feature matching while simultaneously capturing background information that carries category information.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent applies partial action by selectively processing only the most relevant regions of the image rather than uniformly processing the entire image. Using saliency maps to identify and weight only the most important regions reduces the effective computational burden compared to processing all regions with equal intensity, while still achieving high accuracy.

Inventive Principle:

Principle #16Partial or excessive action

3Reliability

If foreground-based approaches are used for feature extraction, then better results are obtained, but important context information from the background is missed

Engineering Contradiction:

Improveobject recognition accuracyVSAvoidbackground context information

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The patent applies local quality by weighting features differently based on their spatial location and relevance. Instead of treating all features equally, the method assigns higher weights to features in regions identified as salient or relevant through saliency maps, while downweighting or ignoring features in less relevant background regions. This resolves the contradiction by maintaining the reliability of key feature matching while simultaneously capturing background information that carries category information.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent makes the feature extraction system universal by enabling it to handle both foreground objects and background context information through a unified weighting mechanism. The same framework that extracts features from salient regions also incorporates background features, making the system capable of processing diverse image content without requiring separate specialized methods.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS8774498B2Modeling images as sets of weighted features

Publication Date: 2014.07.08 GENESEE VALLEY INNOVATIONS LLC

AI summary

An apparatus, method, and computer program product are provided for generating an image representation. The method includes receiving an input digital image, extracting features from the image which are representative of patches of the image, generating weighting factors for the features based on location relevance data for the image, and weighting the extracted features with the weighting factors to form a representation of the image.