The problem of evaluating the aesthetics of photos is considered to be quite complex. Most photographers appear to be more or less skeptical on the ability of automatic aesthetics evaluation, the reasons they give usually refer to the complexity of the nature of photography. Despite this machine learning methods showed efficiency in finding of aesthetically pleasing photos. To apply machine learning for the classification task we need to represent the notions of aesthetics and of certain rules of photography in a formal way. For that purpose a number of various computed image features was proposed and implemented in previous studies, where each of the image features refer to some image characteristics. Still, some of the features may be more useful, the others may show less efficiency for classification. I will review the issues on how to extract the image features, how to evaluate the quality of aesthetics classification and what are the advantages and drawbacks of exiting approaches.