Apparel Recommendation System Using Visual Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users face difficulties in finding complementary apparel items while shopping online, as existing systems struggle to accurately detect and recommend similar items, leading to time-consuming manual searches and inefficient use of system resources.
Innovation Solution
A system utilizing pre-trained machine learning models for image detection and matching, which identifies category-specific features to generate feature vectors and rank candidate items based on similarity, recommending items that are visually similar to those shown with a selected item on a product display page.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If manual search is used to find complementary items, then users can search for items, but it is time-consuming and frustrating
Solution Approach 1:
The system performs preliminary actions by automatically detecting complementary items in product images and pre-computing similarity rankings before the user even searches. The recommendation manager identifies complementary items and ranks them by similarity in advance, so when users view a product page, ready-to-display recommendations are immediately available without requiring manual search effort.
Solution Approach 2:
The patent replaces the mechanical manual search process with an automated image-based detection and matching system. Instead of users manually browsing and searching for complementary items, the system uses image processing to automatically identify complementary products and their features, then substitutes the manual search mechanism with automated similarity ranking and recommendation generation.
2Measurement precision
If traditional search methods are used, then users can search for items, but accuracy is low due to out-of-stock items and sub-optimal search terms
Solution Approach 1:
The system creates visual copies and feature representations of complementary items from product images. By extracting and comparing feature vectors (color, shape, texture, patterns) from images rather than relying on text search terms, the system accurately identifies similar items even when exact matches are out of stock. This visual copying approach bypasses the limitations of text-based search and stock availability issues.
Solution Approach 2:
The patent changes the search parameters from text-based keywords to visual feature parameters. Instead of searching using potentially sub-optimal search terms, the system extracts visual parameters (color histograms, shape descriptors, texture features) from product images and uses these as the basis for similarity matching, significantly improving the accuracy and reliability of finding complementary items.
3Productivity
If comprehensive item detection is performed, then more complementary items can be found, but system resource usage increases
Solution Approach 1:
The system segments the comprehensive detection task into manageable parts by first identifying the main product, then separately detecting complementary items in the same image. It further segments the feature extraction process by focusing on specific relevant features (color, shape, texture) rather than analyzing all possible image properties, reducing computational overhead while maintaining detection comprehensiveness.
Solution Approach 2:
The patent applies partial action by not requiring perfect detection of all possible item attributes, but rather focusing on the most discriminative features needed for effective matching. It performs sufficient (but not excessive) analysis by extracting key visual features that are adequate for identifying complementary items, balancing comprehensiveness with computational efficiency to avoid unnecessary resource consumption.
Data Source
AI summary
Examples provide a system and method for recommending complementary apparel items based on an image of a model wearing an anchor item of apparel. A recommendation manager identifies the category of the complementary item. A per-category similarity definition is used to identify category-specific features used to determine whether a candidate item in the same category as the complementary item is the same or similar to the complementary item. A pre-trained machine learning model is used to calculate feature vectors representing the complementary item and each candidate item. The feature vectors are generated by concatenating feature vector values representing each feature in the plurality of category-specific features for the identified category. The candidates are ranked based on the feature vector values. The highest-ranking candidate items having the greatest degree of similarity to the complementary item are added to a list of recommended items presented to the user.