A Novel Approach for Recognizing Clothes Pattern and Colors for Blind People using DSL-CNN Classifier

Abstract

In this current paper, the presented work based on DSL-CNN Classifier  method for helping and assisting visually impaired people to recognize multiple objects in different class of cloths and colors. This method relies on a novel deep super leering multi-label convolutional neural network (DSL-CNN) for coarse description of all images. The main idea of DSL-CNN is to use a set of linear state vector machines as filter banks for feature required map generation. During the current training phase, the weights of the state vector machines filters are obtained using a forward-supervised deep learning strategy unlike the back propagation algorithm used in standard convolutional neural networks (DSL-CNNs classifier). To handle multi-label detection, the introduced a multi-branch CNNs architecture, where each branch will be used for detecting one object in the cloth images. The current architecture exploits the correlation between the objects present in the image by means of an opportune fusion mechanism of the intermediate different outputs provided by the convolution layers of each branch. The high-level reasoning of the convolutional network is done through binary classification state vector machines for predicting the presence of objects in the image. The currents experiments obtained on two indoor different datasets and through voice the color and its classification will present for the visually impaired people. Final improvements were performed on accuracy, sensitivity, specificity, MCC, and F1-Macro on the proposed model.