Photo Explorer

Photo Explorer

Introduction and related work (doc) (Nov. 5. 2007)

Project proposal

The main goal of this project is to enable easy viewing large pictures on small screens.

Two main motivations for this work are:

1) for cell phones, the screen is very small but the resolution the cell phone has or captures keeps increasing pretty fast, how to navigate a large picture on such a small screen is very interesting.

2) giga-size pictures start to appear (in siggraph papers), people like to see pictures in high resolution, desktop monitor size is far smaller than picture size. How to enable the users to view such big pictures in high resolution is interesting.

The main goals to achieve:

1)      automatically or semi-automatcally identify what the user would like to see (region of interest)

Ø      automatic: for example, detect all the faces in the image (or using pre-trained category-level classifier to detect a particular class of object (e.g. flowers) )

Ø      semi-automatic: the user identifies some examples which he/she is interested for example: flowers and the algorithm finds similar objects in the image using online learning methods (online svm) or fast template matching methods

2) smart navigating among all the region of interests (content aware navigation). One method to view a big picture in high resolution on a small screen is to press the navigation button (either horizontal or vertical) many times(if they are far away) to reach the next region of interest. This is very tedious. The other possibility is to alternating between the low resolution overview picture and high resolution picture. This to some extend alleviates the problem but the interaction is still tedious. We want to smartly navigate among the region of interests. Between regions of interests, the path might be B-spline and transition between two regions can oversee the images in the middle.

If things work well, a by-product would be the capability to search into a picture/a set of pictures (this is motivated by the “search” functionality in word editing software like MS word: given a keyword, find all the exact matches). For high dimensional images, this is a little bit different: find all the similar matches.