This poor woman has been abused by the image processing society in a
huge number of ways. Open the proceedings from any image processing conference, and you
can be sure to find several warped versions of her. A lot of people ask me this question. "What do you really work with?", and so on. Well, I've attempted to give a very short explanation here.
Digital Image Processing is a term that covers a lot of different areas. A common misconception is that this is a new field. Actually, digital images have been created, processed and viewed since very early in this century. But its practical usefulness didn't reach very far before reasonably powerful computers became available in the '70s and '80s.
Obviously, image processing has a lot to do with digital images (images that are discrete in space and amplitude), and the image below is probably the most famous one in the field...
This poor woman has been abused by the image processing society in a
huge number of ways. Open the proceedings from any image processing conference, and you
can be sure to find several warped versions of her.
The picture is actually from the centerfold of the November 1972 Playboy, and Playboy Enterprises has forbidden further
(ab)use of it -- apparently without much success. (The woman, Miss November, is called
Lena Sjöblom and lives outside Stockholm, Sweden. In Playboy, she was called "Lenna
Sjööblom", which is why the picture is often referred to as "lenna".)
This is another image sometimes used in image processing. It originally
comes from the Amiga, and was used in the mid-to-late 1980s to show off the
computer's then-amazing colour capabilities. (The picture is in the Amiga's 4096-colour
HAM mode.)
Some of the main areas/uses of image processing are:
Image compression uses a variety of methods to reduce the amount of data (number of bits) needed to represent an image; for storage or transmission. The excellent JPEG image compression standard is a result of many years' research on image compression techniques. From a 24-bit original, compression ratios of about 10-20:1 can be achieved with virtually no loss of quality. And compressions up to 35-50:1 are often possible with some degradation. The full-sized JPEG of lena above (not the small inlined one) was compressed with a factor 15. Here is a version compressed with a factor 50. The picture shows some artifacting (a "blocky" effect), but it's still quite good.
Of course, the search for better compression algorithm continues. In some of my projects, I am working on interest based compression methods -- where "interesting" parts of the image are detected (automatically or manually), and compressed more lightly than "uninteresting" parts. Such compression techniques might be useful when really high compression ratios are needed.
Here, the task is to "improve" an image from a poor or somehow "degraded" version. Digital images can be enhanced in a variety of ways; to an outsider, it can seem like pure magic sometimes. But there are also things that can not be done. No matter how powerful our computers, no matter how clever our algorithms, we can never "restore" information that isn't present in the original. But it would be fun if it was possible... for instance, here is the output of a "decropping" algorithm, used on the ever-famous Lena image:
Got any suggestions for algorithms to generate this image from the original Lena image at the top of this page? Feel free to send them to me... :-)
For me, this is the most exciting area of image processing (and also the central area of my job). Image analysis is itself a large field, covering scene analysis, computer vision, pattern recognition and other buzzwords. It's basically about teaching computers to see. This is an extremely complicated task, and one that will not be "solved" to any satisfactory degree in our lifetime. We are slowly beginning to create useful applications, but only when strong constraints are imposed to the images, when very simple information is to be extracted, or when a high degree of uncertainty in the extracted information is acceptable. But that's exactly what makes this field so exciting - the challenges are immense.
Per Espen Hagen
Last modified 31.3.96
