Relationships between Image Popularity and Color Emotions in Image Retrieval Systems

Abstract: Colors evoke various emotional feelings which can be described by semantic words such as excitement, harmony, energy, to mention a few. These feelings, evoked either by single- or combinations of colors, are called color emotions. We use previous developed models by Li Chen et al. aimed for estimating values describing color emotions (in this thesis activity, heat, weight and harmony). By calculating histograms, we adopt a previous proposed method by R. Lenz and M.Solli for applying three of the models on ordinary multicolor images. We propose a method using color histogram for applying the harmony model on multicolored images. Out of gathered user data from Picsearchs public search engine, popularity metric for images is approximated. On the basis of a database approximately containing 284 000 images, we analyze the relationship between derived color emotion parameters and approximated popularity values using visual and statistical methods. The experimental results show that individual relationships between different derived emotional values from specific key words and popularity values exist, however, no common relationships that hold for all key words was found.