Image Search Using Pose and Wearing Object Pattern Attributes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image search techniques using pose and appearance information, such as color, face, and body shape, suffer from insufficient accuracy when limited to specific types of information, leading to challenges in identifying desired persons within multiple images.
Innovation Solution
An image processing apparatus and method that incorporates both pose information and detailed appearance information, including specific types of wearing objects and their patterns on particular body parts, to enhance search accuracy by using these characteristics to identify intended images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image search uses only pose information and basic appearance information (color, face, body shape), then the search system is simple and fast, but the search accuracy is insufficient
Solution Approach 1:
The appearance information is segmented into multiple detailed attributes including wearing object type, wearing location, pattern characteristics, and color. This segmentation allows the system to process and match specific visual features independently, improving search accuracy without requiring processing of all image data at once.
Solution Approach 2:
The patent adds new dimensions to the appearance information by incorporating wearing object details and pattern characteristics. This expands the feature space from basic appearance attributes to include contextual information about what the person is wearing and where, enabling more precise matching.
2Measurement precision
If image search uses detailed appearance information including wearing objects and patterns, then search accuracy improves, but the processing time and computational resources increase
Solution Approach 1:
The system extracts and stores detailed appearance information (wearing object type, location, pattern, color) in advance during image processing. This preliminary extraction allows the search function to directly query these pre-processed features without re-analyzing the entire image, reducing search-time computational overhead.
Solution Approach 2:
The patent extracts specific visual features (wearing objects, patterns, colors, locations) from the full image data and represents them as discrete attributes. This extraction separates the essential search criteria from the redundant pixel data, enabling faster comparison during search operations.
3Reliability
If the system distinguishes between wearing objects and their patterns on specific body parts, then the ability to identify intended images improves, but the complexity of information representation increases
Solution Approach 1:
The patent applies local quality by associating wearing object attributes with specific body locations (e.g., 'wearing object on head', 'wearing object on body'). This localization allows the system to distinguish between different wearing objects and their positions without creating a monolithic complex representation, organizing information by spatial and categorical dimensions.
Data Source
AI summary
The present invention provides an image processing apparatus 10 including: an acquisition unit 11 that acquires a search query including pose information indicating a pose of a body of a person and appearance information indicating an appearance of a person; and a search unit 12 that searches for an intended image from a plurality of reference images by using the pose information and the appearance information included in the search query. The appearance information indicates at least one of whether a predetermined kind of a wearing object is worn, whether a predetermined kind of a wearing object is worn on a predetermined portion of a body, whether a wearing object having a predetermined pattern is worn, and whether a wearing object having a predetermined pattern is worn on a predetermined portion of a body.