A System of Clothes Matching for Visually Impaired Persons
Shuai Yuan · 2010 · Proceedings of the 12th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2010) · doi:10.1145/1878803.1878883
Summary
This short paper presents a proof-of-concept computer vision system designed to help blind and visually impaired people determine whether two clothing items match in both color and pattern. The system addresses a common daily living challenge: selecting appropriate clothing combinations without being able to see colors or patterns. Most blind people rely on family members or sighted assistance to manage this task, and at the time of publication no device existed on the market for automated clothes matching. The prototype consists of a sunglasses-mounted camera that captures images of clothing pairs, a mini computer (HP Mini) for processing, a microphone for speech commands, and speakers for audio feedback. The user controls the system through simple voice commands, and receives audio output describing the matching results in four categories: match (both color and pattern match), color match but pattern mismatch, pattern match but color mismatch, and no match.
Key findings
The system uses a three-stage pipeline: color detection and matching, pattern detection, and pattern matching. Color classification operates in HSL (hue, saturation, luminance) color space, creating normalized histograms to identify dominant colors from a palette of ten (red, green, blue, yellow, cyan, magenta, black, grey, and white). Pattern detection uses Canny edge detection followed by morphological processing to determine whether a garment has texture patterns or is uniform in color, using the ratio of edge pixels to total pixels as a threshold. Pattern matching employs Radon transform for orientation-invariant analysis, Haar wavelet features for texture characterization, and gray co-occurrence matrix analysis with six statistical features (mean, variance, smoothness, energy, homogeneity, entropy). The system was evaluated on two databases: a color and texture matching dataset of 128 images and a pattern detection dataset of 45 images. Color classification and matching accuracy reached 99%, pattern detection achieved 100% accuracy, and pattern matching on 133 clothing pairs achieved 85%. The method proved robust to variations in illumination, clothing rotation, and wrinkles.
Relevance
This paper addresses an underexplored area of assistive technology: supporting the daily living task of clothing selection for blind people. While much accessibility research focuses on digital interfaces and navigation, practical tasks like getting dressed appropriately remain significant barriers to independence and social participation. The system demonstrates that computer vision techniques available in 2010 could already achieve high accuracy for color and pattern matching. Since this publication, commercial apps like Be My Eyes, Seeing AI, and various AI-powered clothing identifiers have emerged to address similar needs, validating the practical importance of this research direction. For accessibility practitioners, the work highlights that independence encompasses far more than digital access — it includes the mundane but socially important activities of daily living where technology can reduce reliance on sighted assistance.
Tags: computer vision · blindness · visual impairment · daily living · independent living · assistive technology · color detection · pattern recognition