Emotion Based Image Retrieval

Abstract

Images express opinions and emotions in the process of communication. Since the usage of emotions in images has increased in various scenarios, the area of opinion mining, Human Computer Interaction (HCI), sentiment analysis, fashion design, and design of web pages has recently received a huge burst of interest. Particularly, in social web the ability of identifying different emotions in images might help providing diversification of results, thus proposing different views to users. The emotional content of the picture and the impression it makes on a human is considered and is called EBIR (Emotion Based Image Retrieval). Firstly, feature extraction is done using 3 feature descriptors such as Dominant Color Descriptor (DCD), Color Structure Descriptor (CSD) and Dominant Color Structure Descriptor (DCSD). Then classification is performed by Genetically Optimized Support Vector Machine (GOSVM) and images can be classified into 5 basic emotional categories proposed by Ekman such as Disgust, Fear, Happiness, Sad, and Surprise. Images from the NAPS (Nencki Affective Picture System) database are used for training and testing the images.  Further, the proposed method is compared with the results of Support Vector Machine (SVM) and Least Squares SVM (LSSVM) classifiers. Experimental results demonstrate the efficiency and the supremacy of the approach using measures of precision and recall ratio.