DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
A system of clothes matching for visually impaired persons
1. A system of clothes matching for visually
impaired persons
為視障者自動 搭、配 衣物
學生:楊鴻志
參考文獻
Shuai Yuan
October 2010
ASSETS '10: Proceedings of the 12th international ACM SIGACCESS conference on Computers and accessibility
2. Outline
• Introduction
• Systems and Interface
• Methodology for Clothes Matching
Clothes Color Detection and Matching
Pattern Detection
Pattern Matching
• Experimental Results
• Conclusion and Future Work
3. Introduction
In everyday’s life, people need to find appropriate clothes to
wear. It is a challenging problem for blind people to find clothes
with suitable color and pattern.
In computer vision and image processing research, many methods
were developed for pattern matching and color detection.
However, three issues are critical for successful clothes matching.
The first is the issue of color constancy under people’s perception.
Secondly, shadows and wrinkles are often part of the pattern of
clothes and cause errors. Lastly, many clothes have designs with
complex patterns and multiple colors.
285 million people are visually impaired worldwide: 39 million are blind
and 246 have low vision.
Reference: World Health Organization
4. Figure 1: Match clothes with multiple colors and complex patterns by
using color and pattern information. (a) Three pairs of images of clothes.
(b) Color classification results. (c)Pattern detection. (d) Pattern
similarity measurement results. (e) Final audio outputs.
5. Outline
• Introduction
• Systems and Interface
• Methodology for Clothes Matching
Clothes Color Detection and Matching
Pattern Detection
Pattern Matching
• Experimental Results
• Conclusion and Future Work
6. Systems and Interface
The computer vision-based clothes matching prototype for blind
persons integrates a camera, a microphone, a computer, and
speakers as shown in Fig. 2. The matching results are described
to the blind user by verbal display with minimal distraction of
the user’s hearing sense. The user can control the system by
speech via microphones.
Fig. 2. Prototype hardware and architecture design of
computer vision-based clothes matching aid for blind persons
7. Outline
• Introduction
• Systems and Interface
• Methodology for Clothes Matching
Clothes Color Detection and Matching
Pattern Detection
Pattern Matching
• Experimental Results
• Conclusion and Future Work
8. Methodology for Clothes Matching
Clothes Color Detection and Matching
Our color classifier is based on acquiring a normalized color
histogram for each image of the clothes in bi-conic (hue,
saturation, luminance) HSL space. The key idea is to intelligently
quantize color space based on using the relationships between
hue, saturation and luminance.
Pattern Detection
Based on the color detection results from previous section, if there
is only one dominate color, the input image of clothes has no
pattern. Only for the images with multiple dominate colors, we
continue to check if the multiple colors are caused by texture
patterns.
Pattern Matching
Here we introduce a new approach for pattern analysis using
Radon transform, Wavelet features and Gray co-occurrence matrix.
we employ Radon transform for estimating the orientation information then rotate the image with orientation as
0 degree.
Next, Haar Wavelet transform is applied to obtain features on 3 directions (horizontal, vertical and diagonal).
Finally, the pattern matching is performed based on statistical classification included six features, e.g. mean,
variance, smoothness, energy, homogeneity, and entropy.
9. Outline
• Introduction
• Systems and Interface
• Methodology for Clothes Matching
Clothes Color Detection and Matching
Pattern Detection
Pattern Matching
• Experimental Results
• Conclusion and Future Work
10. Experimental Results
Figure 3. Examples of results for clothes matching. Original image pairs,
wavelet features, gray co-occurrence matrix images and color
detection results are gave respectively.
11. Outline
• Introduction
• Systems and Interface
• Methodology for Clothes Matching
Clothes Color Detection and Matching
Pattern Detection
Pattern Matching
• Experimental Results
• Conclusion and Future Work
12. Conclusion and Future Work
We have developed an efficient system to match clothes with multiple
colors and complex patterns to assist visually impaired and blind
(including color blind) people by distinguish both pattern and color
information.
To handle complex patterns and lighting changes, we combine Radon
transform, wavelet features,and co-occurrence matrix for pattern
matching. Our algorithm for color matching is based on normalized color
in HSL color space.
In addition, we have developed a color classifier to detect multiple colors
including red, green, blue, yellow, cyan, magenta, black, grey, and white.
The algorithm is evaluated by two databases contain clothes images with
a variety of texture patterns, colors,and illumination changes.