Image Retargeting via Saliency Model Pixel Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image retargeting technologies using seam carving algorithms face complex analysis processes and high calculation burdens due to the need to analyze image contents to identify important pixels.

Innovation Solution

Establishing a saliency model to calculate saliency values for each pixel, which determines the importance of pixels, allowing for image retargeting based on these values without analyzing image contents, and using distortion coefficients to adjust offset vectors for improved retargeting.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If seam carving algorithm is used to analyze image contents and identify important pixels, then image retargeting quality is improved, but calculation complexity and processing time increase significantly

Engineering Contradiction:
Improveimage retargeting qualityVSAvoidcalculation complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent transforms the complex image content analysis problem into a simpler parameter-based saliency calculation problem. By defining saliency models with specific parameters (color difference, spatial position, texture features) and calculating saliency values through mathematical formulas, the system avoids complex semantic analysis while maintaining retargeting quality. The transformation from content analysis to parameter calculation significantly reduces computational complexity.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent replaces the mechanical process of manual or algorithmic image content analysis with an automated saliency calculation system based on mathematical models. The saliency model automatically computes importance values for each pixel using predefined parameters and formulas, substituting the need for complex analysis mechanisms while achieving the same goal of identifying important regions.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If traditional seam carving algorithm analyzes image contents to determine pixel importance, then accurate retargeting is achieved, but processing time and computational resources increase

Engineering Contradiction:
Improvepixel importance accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary action by pre-defining saliency models and calculating saliency values for all pixels before the actual retargeting process. The saliency map is computed in advance using efficient mathematical operations on image parameters, allowing the subsequent retargeting to proceed quickly without repeated analysis. This preliminary calculation of pixel importance values significantly reduces processing time during retargeting operations.

Inventive Principle:
Principle #10Preliminary action

3Manufacturing precision

If image retargeting preserves important regions with minimal distortion, then image quality is maintained, but the retargeting process becomes more complex

Engineering Contradiction:
Improveimage qualityVSAvoidretargeting process complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by treating different regions of the image differently based on their saliency values. Important regions with high saliency are preserved with minimal distortion, while less important regions with low saliency can undergo greater deformation. This localized approach to retargeting allows the system to maintain high image quality in critical areas while simplifying the overall process through automated saliency-based region classification.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP2975574B1Method, apparatus and terminal for image retargeting
Publication Date: 2020.03.18 XIAOMI INC
  • EP2975574B1 patent drawingFigure 1~2A
  • EP2975574B1 patent drawingFigure 2B~3
  • EP2975574B1 patent drawingFigure 4

AI summary

The present disclosure relates to a method, an apparatus and a terminal for image retargeting, and belongs to the computer technical field. The method includes: establishing (101) a saliency model for an original image; calculating (102) a saliency value of each pixel point in the original image according to the saliency model; and retargeting (103) the original image to obtain a target image according to the saliency value of each pixel point, the original resolution of the original image and a target resolution of the target image. The apparatus includes a model establishing module, a value calculating module and an image targeting module. Thus, the present disclosure may solve the problem of complicated analysis process and huge calculation amount caused by analyzing image contents of the original image to obtain important pixel points. Consequently, effects of reducing calculation amount may be achieved.