Clothing Search Using Visual Codebooks and Binary Attributes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image search technologies face challenges in efficiently retrieving images of similar clothing due to variations such as geometric deformation, occlusion, size, background, and photometric variability from illumination and pose, which hinder effective categorization and retrieval of clothing in uncontrolled environments.
Innovation Solution
A novel framework for clothing visual search that characterizes clothing using a visual codebook capturing local appearance patterns, integrates face recognition, and employs contextual information and binary attribute classifiers to address variations, constructing a discriminative color codebook and using high-level visual features to reduce noise from image variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image search methods are used to retrieve clothing images, then the search process is simple, but the retrieval accuracy is low due to variations in geometric deformation, occlusion, size, background, and photometric variability
Solution Approach 1:
The patent segments the clothing search problem into multiple stages: first identifying candidate images based on color similarity, then filtering candidates using binary attribute classifiers that examine specific clothing features. This multi-stage segmentation approach improves retrieval accuracy by systematically addressing various sources of variation (geometric, photometric, occlusion) without requiring a single complex system
Solution Approach 2:
The patent transforms the clothing representation from raw pixel data into standardized parameters including color histograms, binary attributes (sleeve type, neckline, pattern), and normalized feature vectors. By changing the parameter representation and using invariant features, the system maintains high retrieval accuracy despite variations in illumination, pose, and background
2Measurement precision
If a comprehensive visual codebook is constructed to capture local appearance patterns, then the clothing recognition accuracy improves, but the processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary actions by pre-computing and storing color histograms, binary attributes, and visual codebook representations for all images in the database during an offline phase. This preprocessing creates ready-to-use feature vectors that can be quickly compared during online search, significantly reducing processing time while maintaining high recognition accuracy
Solution Approach 2:
The patent extracts only the most discriminative local appearance patterns from images to build a compact visual codebook. By selecting and storing only the most informative visual words and attributes rather than complete image data, the system achieves high recognition accuracy with reduced computational burden and faster search processing
3Measurement precision
If binary attribute classifiers are used to filter candidate images, then the retrieval precision improves, but the system complexity increases
Solution Approach 1:
The patent applies binary attribute classifiers that examine specific local clothing features (sleeve type, neckline, pattern presence, button position) rather than analyzing the entire image globally. Each classifier focuses on a particular local attribute, making the system more precise while keeping individual classifier complexity low. The modular local attribute analysis is more manageable than a single complex global classifier
Data Source
AI summary
Methods, systems, and computer readable media with executable instructions, and/or logic are provided for clothing search in images. An example method of clothing search in images can include characterizing clothing within a plurality of reference images using a processor, and characterizing clothing within a query image using a processor. A number of the plurality of reference images having clothing with similar color features as clothing of the query image is identified using a processor. A subset of the identified number of the plurality of reference images having clothing with predefined non-color attributes as clothing of the query image are selected using a processor.