Deseño, construción e validación dun clasificador de imaxes segundo criterios estéticos
- Adrián Carballal Co-director
- Juan Romero Co-director
Defence university: Universidade da Coruña
Fecha de defensa: 22 January 2018
- Fernando Amilcar Bandeira Cardoso Chair
- Nieves Pedreira Souto Secretary
- Javier Rodeiro Iglesias Committee member
Type: Thesis
Abstract
This study intends to propose, and achieve, a computational measure to assess the aesthetic value of an image through different criteria: style, complexity or beauty. In order to obtain this goal, a collection of images are classified and organized following aesthetic criteria. For this purpose different experiences are designed and arranged with the following objectives: 1. Validation of the objective features: Through the identification of authorship of the artistic Works and the discrimination between photographs and paintings. 2. Validation of aesthetic perception of the classifier. 3. Validation of the subjective features: prediction of perception of complexity. In these tasks, There are included an optimal set of metrics to measure the aesthetic complexity; for that purpose, are employed the methods of compression JPEG and fractal, The Zipf’s law, as well as a fractal dimension alongside with filters of border detection Canny and Sobel, on a HSV colour model. Obtained results are compared with other approaches and with the perception by humans. The results show the possibility of predicting the complexity perceived by humans using a set of image-‐metrics related to the complexity; demonstrating the relevance of the estimation metrics based on the compression error in classifications of style and complexity.