Skip to content

Caire - Content Aware Image Resize library

SvenSven www.GSA-Online.de
in Other / Off Topic
Just got to know about this amazing stuff.

Caire is a content aware image resize library based on Seam Carving for Content-Aware Image Resizing paper.

https://github.com/esimov/caire

Meaning it will resize objects in an image, based on heat-map and other things. It is not resizing each part of the image with same metric.

Comments

  • DeeeeeeeeDeeeeeeee the Americas
    What a simple, yet powerful, idea! :o

    I just re-did a WP site yesterday so that the changes I made a while ago now deal with the registered thumbnail images, not file paths, which is better b/c then WP chooses the size out of best available, instead of default (largest), as I had it.

    It's great when it re-sizes and the subject of the image is there. Sad, weird, or lame when it doesn't. (often)

    This is a new idea to me. Is Seam Carving slow? If not, would be great for CMSes. That brings me to a question I had about that...lol
  • SvenSven www.GSA-Online.de
    It is of course way slower than normal resize on percentage or simple new with/height only. Think of the many operations that have to be done to decide what can be stretched and what should be in same shape.
    The samples on the page show the good result.

    However, I never tried it myself. Just find it interesting and a new way of doing this.
  • DeeeeeeeeDeeeeeeee the Americas
    Definitely a cool idea, either way, no doubt! Agreed; very interesting concept.

    This had me wondering if there might be other methods out there already that can help resize intelligently. I've been busy and not as my computer for days, except to briefly so I haven't yet checked if this (or oter methods) already exist.

    So I started looking at tons of images and considering a faster way to calculate what is important as a subject, and how to differentiate backgrounds from  foregrounds.

    It's often not so obvious. :| Same limitation as even such a novel resizing method as Seam Carving.

    But for some images, for example images with just a subject on a plain black or white or mostly plain background, wouldn't comparative counts of color and image data every X or Y rows or columns (to save time calculating) be an easy way to find a fairly constant background, and then just clip the image at the point where the data becomes more disparate row by row, or column by column, with simple statistical calculation.

    For on-the-fly stuff, I see that cutting down the number of calculations is key.

    Definitely not the same method, or nearly as cool as Seam Carving, but it seems that this might work for images with clearly defined backgrounds.
Sign In or Register to comment.