Photo-stream segmentation: Photo Organization

Digital cameras have changed our habits significantly. In early days, we wwaited for ‘kodak moments’ and used our camera thoughtfully. Now we use our camera all the time and take photos generously — photos are becoming our inexpensive secondary memory to preserve our experiences and memories.

A major problem has been how to use all the photos that we take. People take so many photos but how to organize them to retrieve them when needed has become a challenging problem. Flickr, iPhotos, Picasa and such are good starts but lack the flexibility and competence that will help consumers organize these experiences easily for visiting and sharing them. Tags are much hyped so far because who wants to take time to tag? And there is no easy way to tag them.

In computer vision, multimedia, and related communities people realized that most people organize and retrieve photos based on events (time and location). It would be great if automatic techniques could be developed to assign tags automatically to photos when they are loaded on a computer or a web site. SOme research has started in this area. Yesterday Bo Gong defended his doctoral thesis in this area In School of Information and COmputer Sciences on this topic. (Disclosure: I am his advisor.) This work shows how one can take some concrete steps in using context information — coming from EXIF data stored with each picture in modern cameras — in this direction. This is an early work, but is a strong indicator of things that could be done.
A good thing about such research is that it is needed to solve a pain point that the technology has created in the last few years.

4 thoughts on “Photo-stream segmentation: Photo Organization

  1. Kent Tenney

    There are a number of metrics which are easily harvested programatically which
    might be useful if assigned to an image. Given a collection of images, they are
    scanned and the following attributes are determined.

    time delta
    duration between this image and prev, next.

    geometric complexity
    indicated by the degree that the image can be compressed.

    chromatic complexity
    number of discrete colors in image

    average color
    dominant color of image

    I’m sure others will become evident, they share the characteristic of being
    available solely by examining the image data.

    When images are tagged, a Bayesian filter correlates the above attributes with
    tags. I think there’s a real chance that patterns would emerge which, if not
    leading to automation, would provide an improving level of assistance.

    The time delta attribute is valuable in determining batches. Within a batch,
    defined by a set of photos shot within a specified time frame, the tags will
    tend to be the same.


  2. MBA Programs

    Quite frankly, I would rather like to save or organize my photos locally on my machine rather than doing it on internet. Although I don’t feel any harm in doing so but somehow the security concerns are always there.

    There’s no doubt that many websites have very strong security settings as far as images are concerned i.e. you can not just copy the images, but loopholes are always there. I would rather say that arrange for some good photo organizing software (adobe provides such software) and then organize your photos locally.

Leave a Reply