https://www.itu.int/md/T17-SG16-C-0744
With the development of the market economy and the improvement of people's spiritual and cultural needs, the mobile terminal computing photography system not only requires the device to capture high-quality images, but also needs to solve the problem of how to make the image cater to most people's aesthetics. The quality of the image can be measured by objective evaluation, while the aesthetics of the image need to be evaluated by subjective evaluation. In terms of objective evaluation, some well-known standards or systems, such as CPIQ IEEE P1858, DxOmark, VCX, and some single attribute evaluation standards, such as exposure (ISO 12232), dynamic range and noise (ISO 15739), color shading (ISO 17957), and resolution (ISO 12233), etc., have made detailed specifications on related evaluation methods and content, which provide a comprehensive coverage of image quality. By contrast, the standards for subjective evaluation are relatively deficient. Currently, there are only standards such as ITU-R BT.500 and ITU-R BT.1788, which stipulate subjective evaluation methods and conditions for the TV system, but do not point out the attributes of subjective evaluation. Besides, the relevant standards on subjective evaluation of images in mobile terminal are still blank.
Subjective evaluation aims to evaluate whether the image conforms to the possible audience's aesthetics, which belongs to the scope of image aesthetics assessment (IAA). On the one hand, the aesthetic attributes and the processes that affect the final image aesthetics are complex, so the industry lacks a unified definition for it; on the other hand, due to the spiritual and cultural differences among users, even for the same image, there is a different aesthetic evaluation. Therefore, the industry does not know enough about users' demand for aesthetics. By establishing standards, standardizing subjective evaluation and structure of metadata, and using massive annotation data to drive aesthetic evaluation algorithms, we can begin to solve the above problem. The essence of IAA is to quantify aesthetics. It mainly includes two methods: one is to evaluate the overall aesthetics of the image, and the other is to evaluate aesthetics based on multiple aesthetic attributes.
The logical structure of existing metadata lacks interpretability and systematization, and the aesthetic datasets that can be used for training algorithms cover a small scope of annotations, which fail to summarize the key aesthetic attributes systematically. As a result, the generalization ability of IAA algorithms is limited, making them difficult to use in practical applications. The main reason is that the existing standards cannot regulate the content of image aesthetics evaluation and guide the establishment of IAA metadata.
This proposal proposes a method for constructing IAA metadata, it divides the dimensions involved in IAA from the perspective of the role (photographer, camera, viewer), while also providing the aesthetic attributes of each dimension. This proposal attempts to provide support for the following aspects:
1) This proposal can effectively guide the construction of IAA datasets. These datasets can be used to drive aesthetic algorithms, which can assist the ordinary manual or even photography experts to make a comprehensive aesthetic evaluation of images.
2) This proposal can facilitate IAA and improve the cultural orientation and additional value of the product. It provides multiple aesthetic attributes evaluation for IAA, so that IAA can be used in more practical products, such as using IAA to help manufacturers design software that meets the aesthetic needs of users; When taking photos, software can guide users in multiple dimensions to take better images; In the digital media business of mobile terminals, photographic work can be recommended by IAA to better fit the local culture and the aesthetics of the audience.
3) As a sub-standard of the mobile terminal computational photography system series standards, this proposal can fill the gap of standards in the direction of subjective image evaluation, and standardize industry IAA tasks. Moreover, the subsequent standards for each aesthetic attribute will continue to improve this standard and form a standard system with it.
