"Google Photos" newly removed Sharp noise -Most of the images are beautiful -window forests

　Google announced on June 28 that it has integrated two new adjustments, "Sharp" and "Removal", to the "Google Photo" image editing function.You can easily correct the image just by adjusting the slider.

　There are various reasons why photos are blurred and noise, but it is much easier to assume a specific camera device and understand the structure before taking measures.However, that information is not always available.The company has adopted a new approach that does not depend on camera devices.In this case, almost any image can improve quality.

　The first is the removal of the pull-push method (Pull-Push Method).In order to take noise removal locally, it is common to average pixels with local similar structures.However, searching for pixels with similar local structures at a brute force, the larger the size of the image, the greater the calculation cost.Therefore, the noise is reduced for each rough level with the "pull" filter similar to the down sampling.After evaluating to the most coarse level, on the contrary, the "push" stage repeats the "finer" level of noise removal.At this time, efficient calculations can be performed by using similarity information collected by "pull".

　The other is an algorithm for sharpness.The low sharpness, that is, blurred images, are considered as the result of applying "a set of blur (blur) effect" (blur kernel) to the original image.In other words, if you know this "blur kernel", you will be able to get an image without blur in the opposite operation.What is generally called "sharpness" is a process that deteriorates the contour locally without considering the information from other images, and may deteriorate the image quality in some cases.There is no need to worry about processing a blur kernel, but it is difficult to execute with a mobile device.

　However, assuming that the image in front of you is not so blurred that it cannot be repaired, it is said that the estimated blur can be re -applied multiple times and added and subtracted to generate a debris image relatively easily.This method is very fast because the blur itself is applied, and it can withstand a mobile device in practical use.