MIST: A Music Icon Selection Technique Using Neural Network

Thanks to recent evolution of multimedia technology, today we often use computers as music recorders and players. Number of tunes stored in our computers is monotonically increasing, and therefore users often face difficulty while selecting tunes which they want to listen. To solve the problem, we aim visual selection of tunes based on their impression. This paper proposes a technique for automatically selecting icons for music files. Main problem of the proposed technique is matching of tunes and images based on their impressions, where the images are used as icons. The technique first extracts features from images and tunes, and calculate fitness of sensitivity words using Neural Network (NN). Forming multi-dimensional vectors from the fitness values, the technique calculates Euclidian distances from tunes to images. The technique finally selects the closest images as the most matched images for each tune. We think users can visually recognize impressions of a set of tunes, and easily select preferable tunes, if the icons selected by the technique are displayed in folders of file systems.