A picture is worth a thousand words. Pictures tell stories that are naturally drawn to humans. For example, you can click on a multiple image of your pet and choose the one whose pose is hitting instead of the one that has no blur or noise. We do not always choose images for their artistic clarity and there are many times we prefer one image over the other because of the artist's appeal.
Well, what if the AI were to predict which images you find attractive based on aesthetics rather than technology?
Google has developed a deep convolutional neural network in the form of NEMA - Neural Image Evaluation, to improve image rate according to its appeal to users. NEMA can predict what photos users will find attractive and appealing, and then score them on a scale of 1-10 with a high correlation to human perception.
"While the technical quality assessment deals with measuring pixel-level impairments such as noise, blurring, compression artifacts, etc., the aesthetic evaluation captures the semantic plane characteristics associated with the emotions and beauty in the images," Google wrote in detail in its blog post.
So instead of typically categorizing images as low or high quality based on technical factors, NEMA uses deep learning techniques to evaluate images based on aesthetic appeal to humans.
According to NEMA, Google will be beneficial for smart image editing, improving visual quality to increase user engagement, or reduce perceived visual errors in the imaging pipeline. NEMA can also be useful for developing new storage or media sharing techniques. You can see how NEMA has been used to create better aesthetics in photos, a great example of how NEMA scores can be used to enhance photos.
Comments
Post a Comment