Determining Attention Mechanism for Visual Sentiment Analysis of an Image using SVM Classifier in Deep learning based Architecture

摘要

Image (Visual) Sentiment Analysis (ISA) which exhibits the reaction of humans on visual elements for example images and videos, has been an animating and thought provoking problem. ISA demonstrates to apply the fields Computer Vision and Natural Language Processing (NLP) for classifying, extracting, and computing the subjective information in analytical fashion. The accomplishment of existing models can be credited to the progress of solid methodology from Image processing and computer Vision. A large portion of present models were attempted to tackle the problem by highlighting visual features from the complete image or video. Whole image features are the foremost anticipated inputs. For increasing the accuracy of the overall ISA system we proposed a deep learning based model including attention mechanism for consideration instrument for centering local regions of an image determining the required sentiment and adding support vector machine (SVM) in place of soft-max classification layer on deep Convolution Neural Network (CNN). We additionally consider the relevant hash tags of an image to put attention weights on CNN layer indicating by the semantic mapping of image regions and hash tags. Our deep learning based proposed framework is accomplished of spontaneously determining sentimental of specified images and it out spaces existing state-of-the-art approaches to VSA.