Abstract:Automatic photo quality assessment from the perspective of visual aesthetics is a hot research topic due to its potential need in numerous applications. It tries to automatically determine whether a given image has "high" or "low" quality according to the image's visual content. Most existing researches in photo quality assessment predominantly focus on exploring hand-crafted features which may be potentially related to highlevel aesthetic attributes. Most of those features are designed under the guidance of s… Show more
“…This was done not only in the training set, but also in the test set, which makes the problem easier, but further from reality. For example, in [17], we find a comparison between hand-crafted features within the AVA problem domain, where the proposed features outperform the general image descriptors. Nevertheless, this research work only employed 20% of the whole AVA dataset, which simplifies the original task.…”
Automatic aesthetic quality assessment is a computer vision problem in which we quantify the attractiveness or the appealingness of a photograph. This is especially useful in social networks, where the amount of images generated each day requires automation for processing. This work presents Aesthetic Selector, an application able to identify images of high aesthetic quality, showing also relevant information about the decisions and providing the use of the most appropriate filters to enhance a given image. We then analyzed the main proposals in the aesthetic quality field, describing their strengths and weaknesses in order to determine the filters to be included in the application Aesthetic Selector. This proposed application was tested, giving good results, in three different scenarios: image selection, image finding, and filter selection. Besides, we carried out a study of distinct visualization tools to better understand the models’ behavior. These techniques also allow detecting which areas are more relevant within the images when models perform classification. The application also includes this interpretability module. Aesthetic Selector is an innovative and original program, because in the field of aesthetic quality in photography, there are no applications that identify high-quality images and also because it offers the capability of showing information about which parts of the image have affected this decision.
“…This was done not only in the training set, but also in the test set, which makes the problem easier, but further from reality. For example, in [17], we find a comparison between hand-crafted features within the AVA problem domain, where the proposed features outperform the general image descriptors. Nevertheless, this research work only employed 20% of the whole AVA dataset, which simplifies the original task.…”
Automatic aesthetic quality assessment is a computer vision problem in which we quantify the attractiveness or the appealingness of a photograph. This is especially useful in social networks, where the amount of images generated each day requires automation for processing. This work presents Aesthetic Selector, an application able to identify images of high aesthetic quality, showing also relevant information about the decisions and providing the use of the most appropriate filters to enhance a given image. We then analyzed the main proposals in the aesthetic quality field, describing their strengths and weaknesses in order to determine the filters to be included in the application Aesthetic Selector. This proposed application was tested, giving good results, in three different scenarios: image selection, image finding, and filter selection. Besides, we carried out a study of distinct visualization tools to better understand the models’ behavior. These techniques also allow detecting which areas are more relevant within the images when models perform classification. The application also includes this interpretability module. Aesthetic Selector is an innovative and original program, because in the field of aesthetic quality in photography, there are no applications that identify high-quality images and also because it offers the capability of showing information about which parts of the image have affected this decision.
“…Similarly, the low-level and high-level features of images are also applied to the quality evaluation of art works. Dong and Tian [14] considered the importance of the main area, adopted the main background separation, and extracted the color histogram of the main area and the ratio of the main area to the size of the whole image to distinguish the quality of the picture. On the basis of previous studies, Wang et al [15] extracted features such as color, texture, depth of field, complexity, and color harmony and used these features to build an aesthetic classification model, which achieved good results.…”
The average accuracy of the fusion of color harmony and composition features is 75.17%, which is higher than that of a single feature. The classification accuracy of NP-DP-DCNN structure is about 1% higher than that of other methods and 1.77% higher than that of NP-DCNN. Traditional image aesthetic evaluation methods are only effective for specific image sets or specific style images and are not suitable for all types of images. Based on the introduction of the partial differential equation image filtering method, through the parallel supervised learning of aesthetic attribute labels, this paper extracts the global aesthetic depth features, adopts the partial differential equation to evolve the contour C constant, and constructs a convolution neural network. The structure of a convolution kernel learned by using parallel network structure achieves better classification performance. Through the aesthetic evaluation experiment, the overall test accuracy is improved by 0.58% and the average accuracy of the integration of color harmony and composition features is 75.17%, which is higher than that of a single feature. The classification accuracy of NP-DP-DCNN structure is about 1% higher than that of other methods and 1.83% higher than that of NP-DCNN. It has achieved better test accuracy than before in the seven subcategories with discrimination between high aesthetic and low aesthetic images. It has achieved better classification performance than the traditional feature extraction methods in the public dataset CUHK database, and it has excellent aesthetic reference value.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.