Evaluation of Emotions in Artworks using EfficientNet Convolutional Network Integrating the Gram Matrix Modules

This paper explores the possibility of an automatic assessment of emotions in artworks. This is a really important question because it can be extended not only to the emotions of artworks but also to the emotions of the environment around us. The EfficientNet convolutional neural network implementation, integrating the Gram matrix modules, was chosen for the problem. In previous studies, such a solution proved effective for analyzing emotions in images of a general nature. Four image datasets with the known emotion of a particular image were used to find the most appropriate one for the problem. The probabilities with which the neural network assigns emotions to the test art images are estimated. We see significant differences between these probability values. The small probability values obtained allow us to determine which emotions are not inherent in a given image. This is only initial research, as art is perceived subjectively, and the human factor is integral to the solution. However, its results are promising.