Image Processing Demonstration

Introduction - Compression - Plaids - Tutorials

Introduction

When you eyes perceive an image, like a spectacular San Francisco sunset, they receive light which falls on a hexagonal grid of receptors. The brain processes this sampled two-dimensional signal to make sense of the contents of the scene. Similarly, in processing images on a computer, the images are first sampled, but on a rectangular grid. The spacing between grid points sets the resolution in the spatial and frequency domains of the content in the image, just as it does in speech signals. In general, cameras often sample images at much higher resolution than necessary to make out details. The value at each grid point is the intensity at that grid point. The intensity at each grid point is often stored as an 8-bit integer in black-and-white images and in 24 bits in color images (8 bits per ``primary'' color of red, green, blue). For example, here's an example in which we have increased the resolution (decrease the grid spacing) from left to right:
This picture is borrowed from the announcement for ICASSP '96.

Image Compression

One common application of image processing is image compression. In image compression, we seek to reduce the amount of information to represent an image. On Unix systems, file compression is used to compact files without changing the file contents by removing redundant information. This is known as lossless compression. Another approach, known as lossy compression, discards some information but maintains a certain quality. JPEG and MPEG are lossy compression standards for image and video signals, respectively.

Plaids

One approach for lossy compression is to take the image
and decompose it into components
that capture the gross properties of the image, while taking up less space to store. If we add a sufficient number of these components together, we start to get something

We can tradeoff the number of components (i.e., storage size) vs. the quality of the compressed image.

Last updated 10/02/95, Send comments to (Mailbox)ble@eecs.berkeley.edu