+ Ötletistálló +

Erre igazán büszkék vagyunk! close

+ MY-X DVD +

A MY-X team-et is képviselve BME ETDK-szekciógyőzelem! Irány az OTDK!

A MY-X team új tagjai (a BPROF-képzés I. éves Hallgatói) ETDK-I. és II. helyezést, ill. 2 OTDK-továbbjutást értek el!

Angolul oktató avatárok a QuILT-ben!

Már angolul is gyűlnek az egyre minőségibb publikációk a nemzetközi képzések keretében!

A Liebig-elv felfedezése a közlekedésben!

Sikeres részvétel a III. Gazdálkodás és menedzsment tudományos konferencián, avagy versenyképesség és innováció, mint kulcsszavak és a 2DM-játék!

Díjak az MMO 2018 konferenciáról (a szekció legjobb előadása a 2DM-ről, a szekció legjobb publikációja a Rosling-animációkról)!

Last modified: 2015.VII.19.14:08 - MIAÚ-RSS
A MIAÚ immár 20­+ éve áll a Köz szolgálatára!

Robot-Esthete I./II.

Leading article: 2021. February (MIAU No. 270.)
(Previous article: MIAU No. 269.)

Keywords: arts, automation, chained similarity analysis, anti-discriminative modelling, optimizing, automation, artificial seeing, artificial intelligence

Abstract: The Vitruvian man of Leonardo Da Vinci is a kind of evidence of the specific ratio (divine proportion, golden ratio) being available in the nature (c.f. phi=1.1618). Parallel, the mathematics of the beauty has to derive a scale for robot-eyes. This scale should be capable of estimating/describing the difference between two pictures (here and now between a randomized pixel-set and a real nonfigurative painting) in a context-free way. The scale has to support the ranking of the beauty of pictures. It is therefore possible to derive beauty and its inverse, or even a good-better-best distinction between pictures. The kind of artificial intelligence-based beauty-definition has to lead to a positive Turing-test – without any training, without human evaluation of pictures in advance. The robotaesthetic-expert (the robot-eye) is capable of sensing/estimating the ratio of the visual conception and the coincidence. This beauty-scale could be derived from the context-free risk-definition of the similarity analysis where risk-free is a data pattern where each data position can be derived from the rest. The same logic can be used for the definition of the harmony/beauty. A kind of side-effect is a visual risk-map. This map gives red alert signals if the interpretability of a part of the picture is less than the interpretability of the neighbourhood. The risk-map can be used by artists to improve the compositions – like sounds of a less trained singers can be optimized based on automated adjustments processes.
The I. part about the robot-aesthete presented the theoretical backgrounds concerning the mathematics of the beauty as such. This part (II) demonstrates a Turing-test-process, where 15 random-selected, but nonfigurative pictures got evaluated by a random set of human beings (by 36 people) on an artificial beauty-scale (1<10) in a separated evaluation process for coloured and greyscaled versions. The analysis of the human evaluation should deliver a statement for the question: Is it possible to create a scale for measuring the randomness of the human answers concerning each picture? This quasi classic (statistic-oriented) challenge should be solved based on similarity analyses – based on the same AI-engine what will be used for further analyses concerning the beauty as such. This randomness-scale will let derive: which picture could be evaluated in the most robust way (it means: less random-like) and which picture generated the most random-like human evaluations. Therefore, we will also know, which pictures are under-norm, over-norm or even norm-like concerning the randomness of the human evaluation. All these will be derived without using the logic of significant differences. Based on the raw human evaluation, a kind of aggregated (similarity-oriented) beautyindex will also be calculated. Parallel, a naïve robot-eye will also be constructed being just capable of estimating grey-scale-values for each pixel based on the neighbourhood (it means: on 3*3 pixels). The estimated pixels and the real pixels lead to differences, where the more beautiful is a picture, the less is the volume of the differences (describe through a lot of statistical attributes). The results: 3 of the analysed pictures could be evaluated as valid objects because of random-like human evaluations and/or modelling anomalies. There is one single picture where the naïve robot-eye and the human beauty-conceptions lead to the same conclusion: this single picture is an under-norm-object concerning the pattern-oriented beauty-definition. Therefore, it is not a trivial successful Turing-test – but the success is more depending on the human beings as the naïve robot-solution.
Part I. (DOC) + Part II. (DOC) *** Part I. (PDF) + Part II. (PDF)

Please, send Your comments per email!