Classification of Clothing Using Convolutional Neural Network

Abstract

This paper presents classification of an image as shirt, T-shirt or trouser for a specific objective by training a convolutional neural network (CNN). The classifier implemented is a significant component of the assistive instrument developed to help people with dementia become more independent with dressing. The work presented in this paper brings out tuning the hyperparameters of the CNN used in the system. A dataset was prepared for the three classes of clothing by capturing the images, pre-processing and labelling the images. Data augmentation was performed on a subset of the original dataset to reduce the overfitting problem. A standard architecture was chosen with convolution, max-pooling and dropout filters which help in dimension reduction, thus enabling faster training of the model. Upon evaluation of the model on the testing dataset, an accuracy of 93.31% was achieved. In order to describe the performance of the classification model, a confusion matrix was plotted.