Abstract

We demonstrate the use of DeCAF and Improved Fisher Vector image features for clothing texture classification, with a focus on aiding visually impaired individuals in selecting their attire. Choosing clothes based on color and texture is a daily problem for visually impaired individuals. This work is a preliminary attempt to alleviate the problem using a finger-mounted camera and state-of-the-art classification algorithms to identify clothing textures. In evaluating our solution, we used a NanEyeGS camera, small enough to mount on a finger, to collect 520 close-up images across 29 garments, referred to as the HandSight Color Texture Dataset (HCTD). Secondly, we contribute evaluations of these state-of-the-art recognition algorithms applied to our dataset - achieving an accuracy exceeding 95%. Throughout this article, we review prior research, evaluate our current solution, and outline future project directions.

Statistics