With the advent of the multi-device era, cross-platform adaptation of images and videos has become an urgent need. How to automatically and efficiently adjust image size to adapt to different screen sizes and maintain the best display effect has become a research hotspot in the field of image processing. A research team from the University of Sharjah in the United Arab Emirates has provided an innovative solution based on deep learning, which can automatically predict the optimal size of images and select the most appropriate redirection technology, effectively reducing information loss and improving user experience.
With the rapid popularity of digital devices, how to perfectly adapt images and videos to various screen sizes has become an urgent problem to be solved. A research team from the University of Sharjah in the United Arab Emirates recently published a study using a deep learning model to develop a new technology that can automatically predict the optimal size of images to achieve seamless display between different devices.
The core of this research is the use of transfer learning technology, using deep learning models such as Resnet18, DenseNet121 and InceptionV3. Researchers said that although there are many existing image retargeting technologies, they often cannot automatically adjust the image size and still require manual intervention. This results in images that may appear cropped or distorted on different screens. Therefore, the research team hopes to find the best image redirection method through automated means to reduce information loss and maintain image quality.
To achieve this goal, the researchers constructed a dataset containing 46,716 images of different resolutions involving six categories of retargeting techniques. Through experiments, they used category information as a third input while encoding resolution information as an additional channel of the image. After evaluation, the results show that their method achieves the best F1 score of 90% in selecting appropriate redirection techniques, indicating the effectiveness of this method.
The research team believes that deep learning can automatically extract image features and effectively capture complex relationships, thereby making the classification of image retargeting methods more accurate. While a commercialization timeline for the new technology has not yet been revealed, they highlighted the need for further research to develop models that fully automate the selection of the best technology and retargeting of images. In addition, they plan to expand the dataset, adding more samples and redirection methods to improve the accuracy and adaptability of the model.
This research provides new solutions for the field of image processing, and we look forward to achieving more efficient and intelligent image redirection in the future.
Paper: https://ieeexplore.ieee.org/document/10776979
Highlight:
The research team developed automatic image redirection technology based on deep learning that can seamlessly adapt to different screens.
Models such as Resnet18, DenseNet121 and InceptionV3 are used to significantly improve the accuracy of image processing.
By expanding the data set and further research, the team hopes to achieve a more comprehensive automated image processing solution.
This research result provides a new idea for solving the image adaptation problem, and its high accuracy and automation bring new possibilities for the development of future image processing technology. The research team’s subsequent efforts, especially the expansion of the data set and the improvement of the model, will further enhance the practicality and popularity of the technology.