Focus: This semester we plan to discuss Applications of Computer Vision that are of interest in the Internet: Pictorial Information Retrieval, Image Compression, Authentication (CAPTCHA), etc.
Pictorial Information Retrieval: This is an old, and still, unsolved problem: the recovery of information from a pictorial data base using visual queries. For example, we may submit a sketch and ask to retrieve all images that contain shapes similar to the sketch. The current state of the art relies on text queries for pictorial databases where each image has a textual description attached to it. This method works well if the text descriptions are available but that requires an expensive preprocessing of the data base. We would like to,say, submit a picture of a person and retrieve all images where the person appears. This is not a simple image matching problem because different images may have been taken from different distances, different illumination, and different viewpoint. - A Practical Application: Back in 1998 reporters were searching the images of the data bases of News Organization for pictures where both president Clinton and Monica Lewinsky appeared. If the technology we are discussing was available then, the task would have been fairly easy. Instead it required large amounts of human labor.
Image Compression: It will be helpful to have an automatic way to select the type of compression (and file format) to be used for a given image without the need for human experimentation. (This will be useful for web page developers.)
Authentication: Some web sites like to provide accounts only to human users and keep out programs. One way to achieve this is to display an image and request that the person registering type in the interpretation of the image. Of course a program may contain image analysis algorithm and be able to overcome this obstacle so the design of such images is nontrivial.
Watch this space for more details.
Updated August 25, 2004