Face-Aware Image Cropping for Multi-Face Salience

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for reducing the size of pictures or videos often result in unusable parts, especially when dealing with multiple faces, as they fail to appropriately define a cropping window that highlights the most salient areas.

Innovation Solution

A method that detects faces in an image, determines weights based on face size, and modifies the cropping window to be centered on a bounding box, with the option to order and prioritize faces by weight for multiple faces, ensuring the cropping window captures the most important visual elements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If subsampling is used to reduce picture size, then the number of pictures displayed simultaneously increases, but some picture parts become unusable as they are too small

Engineering Contradiction:
Improvenumber of pictures displayed simultaneouslyVSAvoidpicture quality
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The patent extracts the most salient parts of the picture (containing faces) and places them in a dedicated salient area of the reduced picture, while other parts are placed in the remaining area. This extraction principle allows the most important content to be preserved at higher quality while enabling simultaneous display of multiple pictures.

Inventive Principle:
Principle #2Taking out (Extraction)

2Manufacturing precision

If cropping is used to highlight salient areas, then picture quality improves, but the method fails to define appropriate cropping window in the presence of multiple faces

Engineering Contradiction:
Improvepicture qualityVSAvoidhandling multiple faces
Core Design Contradiction:
Manufacturing precisionVSAdaptability or versatility

Solution Approach 1:

The patent segments the reduced picture into two distinct areas: a salient area for displaying the most important content (faces) and a remaining area for other picture content. This segmentation allows different handling strategies for different parts of the image, enabling proper management of multiple faces while maintaining overall picture quality.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different quality levels to different parts of the reduced picture: the salient area receives higher quality treatment with dedicated face placement, while the remaining area uses standard placement. This local quality differentiation ensures that faces are always clearly visible while maintaining efficient use of the reduced picture space.

Inventive Principle:
Principle #3Local quality

3Manufacturing precision

If face detection and weight-based positioning is implemented, then the cropping window captures the most important visual elements, but the processing complexity increases

Engineering Contradiction:
Improvevisual element capture accuracyVSAvoidprocessing complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent performs face detection and weight calculation in advance before generating the reduced picture. By pre-identifying face locations and calculating their importance weights, the system can efficiently place faces in the salient area during the reduction process without adding significant complexity to the final output generation.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3114687B1Method and device for processing a picture
Publication Date: 2021.05.05 THOMSON LICENSING SA
  • EP3114687B1 patent drawingFigure 1~2
  • EP3114687B1 patent drawingFigure 3
  • EP3114687B1 patent drawingFigure 4

AI summary

A method for processing a picture comprising at least one face is provided. The method comprises: - obtaining (S10) a cropping window in the picture; - processing (S18) the picture by cropping the picture part delimited by the cropping window; wherein the method further comprises detecting (S12) the at least one face, determining (S14) a weight for the detected at least one face and modifying (S16) the position of the cropping window in the picture based on the weight, wherein the weight is determined at least based on the size of the corresponding detected face.