| dc.description.abstract |
Shooting a good picture requires a lots of photography skills and good equipment. Thankfully, modern technology can make up for non-professionals. If someone taking a picture using the wrong exposure setting or the camera is not good enough to reproduce the scene, the photo produced would be in low contrast and no color. Thankfully, image enhancement technique can retrieve the lost detail and color information. However, image enhancement cannot deal with image that has lost almost all detail information. HDR technique is a solution to this kind of problem.
There are many ways to generate a HDR image. They are mainly about how to do tone reproduction or how to do the exposure fusion. Others are focus on removing ghost effect. However, there are rarely studies which are related to people′s feeling and the preferences to the HDR images. In other words, we have few research about what kinds of images (HDR images) are the good images depend on subjective feeling while we have many ways to generate all kinds of different HDR images. Hence, in this paper, we want to find out whether there is an index can reflect subjective feeling to HDR image. The method we proposed is call S_fpg. The foreground pixel used in image enhancement to evaluate the performance is used to find out the detail region of a HDR image because image enhancement and HDR are basically the same thing-retrieve the detail information from scene to image. Once finding out the foreground, the saturation measure is evolved to see the colorfulness of HDR image.
A subjective study about HDR images preferences survey is conducted using online survey system. We conduct two kinds of survey to get the preferences data. One is choose-the-best-one and the other is two alternative forced choice. The reason we do choose-the-best-one test is that the conventional mean opinion score survey cause too much cognitive load to the testers. Furthermore, the score testers graded doesn’t have a standard. Ranking the preference image is another method. However, the testers have difficulty in ranking the image that they don’t like. Hence, we change the survey method to choose-the-best-one which significantly reduce the cognitive load to testers. Two alternative forced choice is suggested to be more intuitive and hence, we conduct another survey using 2AFC.
Both of two results are consist with our index. The index is validated by correlation coefficient with subjective test using Spearman′s rank correlation coefficient and Pearson correlation coefficient. The results show that our method is highly correlated with subjective preferences comparing with other objective measure metrics. | en_US |