January 20, 2025
The GIST Editors' notes
This text has been reviewed in response to Science X's editorial course of and insurance policies. Editors have highlighted the next attributes whereas guaranteeing the content material's credibility:
fact-checked
peer-reviewed publication
trusted supply
proofread
AI-driven picture retargeting: Predicting dimensions for seamless show throughout completely different units
Utilizing deep studying fashions, scientists at Sharjah College have designed strategies to routinely predict appropriate picture dimensions to suit completely different screens or units much more effectively and successfully than present applied sciences used for picture cropping and resizing.
The deep studying fashions the scientists suggest of their analysis are primarily based on switch studying units, corresponding to Resnet18, DenseNet121, and InceptionV3. They declare their fashions can predict the right dimensions for the enter photos with particular decision.
The work is printed within the journal IEEE Entry.
With the appearance of computer systems, picture retargeting has grow to be a extensively practiced approach. It primarily adjusts the scale of an enter picture by concurrently preserving its visible qualities and particulars as it’s being resized to suit onto a wide range of screens or units.
The rise of digital units, together with smartphones, tablets, and computer systems, has made it essential to make dynamic changes to the sizes of photos and movies to accommodate the precise show necessities of every machine.
Numerous picture retargeting strategies are at the moment obtainable and reasonably priced, and the authors point out cropping (CR), Scaling (SCL), seam carving (SC), warping (WARP), Scale-and-Stretch (SNS), Multi-Operator (MULTI), amongst others.
Nevertheless, they preserve that obtainable strategies fall in need of the flexibility to regulate a picture's dimension by themselves with out direct human management, "since completely different screens characteristic completely different facet ratios that would trigger photos not optimized for that display to grow to be cropped or distorted."
They throw mild on the "hole in automating the choice of the very best retargeting strategy primarily based on a picture and the goal decision," saying that their "analysis makes an attempt to bridge this hole and goals to construct a mannequin to find out which approach finest retargets a picture to reduce data loss and protect high quality."
The authors imagine automation is the easiest way to focus on a particular picture with a particular dimension. Because of this, they suggest deep studying fashions primarily based on switch studying, and so they use switch studying units corresponding to Resnet18, DenseNet121, and InceptionV3 as instruments with the flexibility "to foretell the acceptable retargeting technique for the enter picture with a particular decision."
Switch studying employs machine studying strategies by way of which fashions constructed particularly for a specific job are suitably adjusted to suit one other project. Resnet18, DenseNet121, and InceptionV3 are deep studying fashions designed to carry out a wide range of duties that embody understanding picture particulars and construction, picture recognition and classification in addition to object detection and picture segmentation.
The authors used a dataset of 46,716 photos of various resolutions from numerous retargeting strategies belonging to 6 classes. They are saying they performed experiments "with fashions the place the class is fed as a 3rd enter and with the resolutions encoded as an additional channel within the picture.
"Moreover, the fashions are evaluated with numerous analysis metrics. The outcomes demonstrated the effectiveness of the proposed strategy for choosing the right retargeting approach with a finest case F1 rating of 90%."
The authors tout their approach as "optimizing" because it renders the prediction of picture concentrating on duties as efficient and practical as doable. They write, "By optimizing photos to suit completely different display sizes and facet ratios, we will guarantee they show accurately on numerous units and look good no matter display dimension or facet ratio variations.
"Deep studying has rapidly grow to be one of many go-to strategies for classifying picture retargeting strategies since its capabilities permit it to routinely extract options from photos and successfully seize all complicated relationships."
The authors current what they describe as "novel switch studying fashions for the automated identification of picture retargeting strategies. We used a number of fashions, corresponding to base CNN, Resnet18, DenseNet121, and InceptionV3. The fashions are assessed utilizing a wide range of analysis measures, and the outcomes present how efficient the fashions are for selecting the very best retargeting technique."
The authors don’t point out once they anticipate their new picture retargeting strategies to be commercially obtainable, however they spotlight the necessity for additional analysis to "develop a mannequin that chooses the very best approach and retargets the picture to the required decision in a completely computerized strategy."
As well as, they "plan to increase the annotated dataset with extra samples and extra retargeting strategies to supply a extra exact and correct mannequin that may generalize to many use instances."
Extra data: Mohammad Alsmirat et al, Supervised Deep Studying for Very best Identification of Picture Retargeting Strategies, IEEE Entry (2024). DOI: 10.1109/ACCESS.2024.3510675
Journal data: IEEE Access Supplied by College of Sharjah Quotation: AI-driven picture retargeting: Predicting dimensions for seamless show throughout completely different units (2025, January 20) retrieved 20 January 2025 from https://techxplore.com/information/2025-01-ai-driven-image-retargeting-dimensions.html This doc is topic to copyright. Other than any honest dealing for the aim of personal examine or analysis, no half could also be reproduced with out the written permission. The content material is supplied for data functions solely.
Discover additional
Environment friendly machine studying: Predicting materials properties with restricted knowledge 0 shares
Feedback to editors
