IMAGE AESTHETICS EVALUATION THROUGH AI ALGORITHMS

The evaluation of image aesthetics through automation has become a significant research issue because of the fact that social media, photography and creative industries are expanding at a very fast rate compared to digital imagery. Contrary to typical vision exercises, aesthetic evaluation is multi-dimensional, subjective by nature, and perceptual, semantic, and emotional. In this paper, the analysis of image aesthetics evaluation based on artificial intelligence algorithms is carried out in detail, starting with the traditional methods of the evaluation based on the handwritten feature and concluding with the advanced deep learning ones. We compare convolutional and vision convolutional neural networks, as well as hybrid networks, and point out their advantages in the local visual quality of modeling and the global compositional structure. In order to overcome the inconsistency in the human judgment, the research focuses on subjectivity-sensitive learning, using the distribution-based annotation and the model of the pairwise preferences. An aesthetic scoring system based on the combination of regression, probabilistic distribution learning and ranking objectives is addressed in terms of implementation on the system level. Through experimental analysis and discussion, it has been shown that hybrid models are stronger, easier to interpret and closer to human perception. The results prove the relevance of holistic learning of features, uncertainty modeling, and explainable decision-making to reliable and human-congruent aesthetic evaluation systems.