Abstract:
People take for granted how much they use their vision when they shop, and the current online tools are not sufficiently supported or developed. We explore approaches to provide a more accurate and pleasant user experience. Convolutional neural networks (CNNs) and their application of image processing are at the center of the solution to this problem. The algorithms discussed include CNNs in general, fully convolutional neural networks, and deconvolution networks. A comprehensive explanation of basic neural networks is given along with a detailed overview of convolutional neural networks. The application of these networks is seen in the three main papers discussed. A paper by Stangl et al. (2018) looks at an application called Browse With Me that uses a fully-convolutional network to identify 10 different items of clothing. They create an interactive environment for a visually impaired user to control the system with their voice and gain the information they want. In a paper by Kim et al. (2018) they use a deep learning, multimodal algorithm that identifies categories of graphs in popular media articles. Although these papers make great strides in accessibility research, the fully convolutional network is unable to identify the details needed to satisfy the visual impaired user. Deconvolution neural networks can be used to investigate whether this algorithm is better suited to satisfy the needs of visually impaired people to participate in the world's increasingly important e-commerce platform (Noh, 2015). Alternatively, implementing a fully convolutional and deconvolution network hybrid into the Stangl et al. (2018) system provides an opportunity to test and determine which algorithms can serve visually impaired shoppers best. The ultimate goal is to include everyone in our society that increasingly relies on an online world.
Except where otherwise noted, this item's license is described as http://creativecommons.org/licenses/by-nc/4.0/