Aesthetic Image Statistics Vary with Artistic Genre
Round 1
Reviewer 1 Report
By means of a rather sophisticated statistical analysis the author shows how different image statistics relate to the aesthetics of different paintings categories. This paper is very pertinent to this SI and I have no reservations in recommending it for publication.
I have just minor comments that the author may wish to consider.
It is written that Mondrian agrees with Leger but the link between the two quotes is a bit opaque to me, maybe the author would like to make it explicit? I would make it clearer from the beginning that the paper focuses on visual aesthetics and visual artworks, not other senses. There seem to be some problems with referencing with authors names being in brackets even when they are the subjects of the sentences. Somehow related to point 2, at the end of page 2 it is written “multi-dimensional nature of artworks”. But here the author refers merely to visual characteristics. The term multi-dimensional sounds ambiguous in this context: a painting representing a human face is obviously different from a painting representing a landscape, but this difference is not commonly thoughas a difference in a “dimension”. I would add a brief rationale of why these three specific image statistics were computed. I would explain PLSR a bit earlier in the paper, the first time it is mentioned, at the beginning of page 3. The analysis used the categories adopted in the JenAesthetics database. I would add a sentence to justify why these categories are good for this specific test. This is particularly important considering that the main finding of the paper is that different image statistics appear to be important for these categories. Where there any paintings including people together with landscapes? Or animals together with people, etc? If so, how were those classified?
Author Response
It is written that Mondrian agrees with Leger but the link between the two quotes is a bit opaque to me, maybe the author would like to make it explicit? I would make it clearer from the beginning that the paper focuses on visual aesthetics and visual artworks, not other senses.
Response: I have re-worded lines 20-21 to improve clarity.
There seem to be some problems with referencing with authors names being in brackets even when they are the subjects of the sentences. Somehow related to point 2, at the end of page 2 it is written “multi-dimensional nature of artworks”. But here the author refers merely to visual characteristics.
The term multi-dimensional sounds ambiguous in this context: a painting representing a human face is obviously different from a painting representing a landscape, but this difference is not commonly thought as a difference in a “dimension”.
Response: I have removed references to “multi-dimensional” in lines 45-59, and corrected all instances of rogue parentheses.
I would add a brief rationale of why these three specific image statistics were computed.
Response: I have added a brief rationale at lines 63-64.
I would explain PLSR a bit earlier in the paper, the first time it is mentioned, at the beginning of page 3.
Response: I have adjusted text ordering a little; PLSR is first mentioned in two places on page 2 (lines 61, 84)
The analysis used the categories adopted in the JenAesthetics database. I would add a sentence to justify why these categories are good for this specific test. This is particularly important considering that the main finding of the paper is that different image statistics appear to be important for these categories. Where there any paintings including people together with landscapes? Or animals together with people, etc? If so, how were those classified?
Response: I have added a sentence to explain why the categories were retained at lines 104-105.
Reviewer 2 Report
This is a short and concise look into the relationship between certain statistical image properties and their aesthetic evaluations. While the association between these measure was very weak for the entire set, there were stronger associations and better partial least squares regression fits for the particular artistic genres. While the analysis seem comprehensive (including 476 artworks split into different artistic genres) with careful and precise calculations of image statistics (spectral slope (SL), fractal dimension (FD) and entropy (E)), I am not particularly convinced by the conclusion that the different statistical characteristics are relevant to different kinds of scenes, at least for some of the artistic genres studied.
It goes without saying that different scenes contain somewhat different distributions of various image properties ranging from spatial frequency to orientation, color, among many others. But, it is not at all clear how is this relevant in this particular context (aesthetic ratings).
In particular, there are no specific discussion of the direction of association between different image statistics and aesthetic ratings. All correlations are reported as Coefficients of Determination (r2x100), it is not clear whether the relationship with the particular image statistics and aesthetic evaluation is positive and negative.. The direction of association surely must play a role if one is claiming the importance of the reliance on the selective range of image statistical properties (for example with respect to the natural scene statistics).
Also, while Figure 1 does seem to indicate a certain degree of overlap between statistics of different painting genres (except for the abstract paintings), there are some obvious differences between different genres. Has there been any formal statistical analysis of the differences in these distributions?
As a final point, the introduction seems to omit quite a substantial number of papers which show a strong relationship between the parametric manipulation of statistical image properties and aesthetic evaluation... These studies do not use correlations but do seem to show quite a robust effects of the variations in the spectral slope and fractal dimension on the aesthetic ratings.
Author Response
In particular, there are no specific discussion of the direction of association between different image statistics and aesthetic ratings. All correlations are reported as Coefficients of Determination (r2x100), it is not clear whether the relationship with the particular image statistics and aesthetic evaluation is positive and negative.. The direction of association surely must play a role if one is claiming the importance of the reliance on the selective range of image statistical properties (for example with respect to the natural scene statistics).
Response: The directions of associations were already mentioned in the results (lines 186, 191, 195-197, 201-202, 205-206, 209-210, 213-214, 218). I have added comments on the direction of the associations to the Discussion (lines 258-262, 268-272, 278-283, 292-293).
Also, while Figure 1 does seem to indicate a certain degree of overlap between statistics of different painting genres (except for the abstract paintings), there are some obvious differences between different genres. Has there been any formal statistical analysis of the differences in these distributions?
Response: I have performed additional statistical analyses, and added the results of these analyses to the Discussion at lines 244-253.
As a final point, the introduction seems to omit quite a substantial number of papers which show a strong relationship between the parametric manipulation of statistical image properties and aesthetic evaluation... These studies do not use correlations but do seem to show quite a robust effects of the variations in the spectral slope and fractal dimension on the aesthetic ratings.
I have made relevant changes to the text in the Introduction at lines 31-32, 41-43. Relevant literature is also cited in the descriptions of the statistics (lines 70-71, 76, 82).
