Zooming

In the last years, we observed a large diffusion of digital images in the everyday life, and electronic devices able to capture digital images, such as digital cameras, PDA, cell phones, etc., have become very popular. Moreover, streaming video websites such as YouTube have increased the possibility to share images and videos. The diffusion of digital images has also a social impact, since new products have become available to meet new requests of the community. For instance, videosurveillance permits to keep big areas under control, improving the public safety, and many medical applications based on image processing increase the possibilities of the medicine.

Often, however, the quality of the images is not as good as desired. In particular, the resolution of the image has to be limited due to the cost of the acquisition device and to reduce the amount of data that has to be saved. Therefore, in many situations it is not possible to detect or recognize small details. This is a problem both for the human observers and for the computer vision algorithms applied to the images.

The procedure to enlarge the size of the image is known as zooming. It requires an interpolation of the available data of the lower-resolution image, in order to estimate the missing pixels and obtain a higher-resolution image. As example, zooming finds application in streaming video such as with YouTube, which often stores video at low resolutions (e.g. the 352 x 288 pixels CIF format), since users often wish to expand the size of the video to watch at full screen with resolutions of 1024 x 768 or higher. Other applications are related to law enforcement and surveillance, where usually poor quality low-resolution images are obtained from typical low-quality commercial video cameras or snapshots acquired in a non-ideal imaging environment. From these poor low-resolution images it is required to have a better quality high-resolution image with clearer details that enhance its use as evidence and to support crime scene analysis. Often a zooming is necessary also in aerial and satellite imaging where it is useful to magnify a specific region of interest or to view details beyond the limit of the satellite imaging system resolution. Another application of a zooming algorithm is in medical imaging in order to let the phisicians to analyze also the details of the image and to make a better diagnosis.

The most common methods used in practice, such as the nearest neighbor replication or the bilinear and bicubic interpolation, require only a small amount of computation. However, these simple methods often produce images with various problems along object boundaries, including aliasing, blurring, and zigzagging edges. Motivated by these drawbacks, various adaptive interpolation algorithms have been proposed to adapt the estimation to the local behavior of the image and reduce the artifacts near the edges.