• Luca Marchesotti , Marco Bressan
IS&T/SPIE s International Symposium on Electronic Imaging , Image Quality and System Performance V, 27–31, San Jose, California, USA, January 2008
Assessing the perceptual quality of pictures still remains a diffictult task even for humans. This is true, especially when there are many interesting regions to look at (e.g. sea and foreground subject) or when the differences among the pictures are subtle. Despite that, trends in user preference do exist and they can be a valuable source of information for designing enhancement algorithms. However, a major problem is to assess preference trends and to translate them in an algorithm with a formal methodology. The approach that we describe in this paper proposes a multi-step solution. In the first instance we relate the space of possible enhancement sequences (intended as chain of enhancement algorithms) to the content of the image and them reduce the number of sequences through an iterative selection penalizing the sequences that produce artifacts or that generates close results. We then present the user with pairs of images enhanced with the various sequences and we ask to select the best in each comparison. Finally, we perform a statistical analysis of users votes through a statistical method. Preliminary results show preference for saturated and colorful sea and sky and desaturated snow.
Report number: