Browse Prior Art Database

Compression of Portrait Image Data

IP.com Disclosure Number: IPCOM000118039D
Original Publication Date: 1996-Aug-01
Included in the Prior Art Database: 2005-Mar-31
Document File: 2 page(s) / 75K

Publishing Venue

IBM

Related People

Chevion, DS: AUTHOR

Abstract

While digital coding methods are available for coding images such as portraits, they typically result in a large amount of data. There is a growing need for the storage of such images on credit cards and smart cards where the amount of storage available is severely limited. By combining a selective compression approach to critical image features, such as EYES, NOSE and MOUTH, with a particularly effective compression algorithm as described here, it is possible to obtain substantial additional compression without materially affecting the quality of the image. A portrait image which would typically occupy 1500 bytes using waveform encoding can be effectively compressed to about 250 bytes using the disclosed technique.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Compression of Portrait Image Data

      While digital coding methods are available for coding images
such as portraits, they typically result in a large amount of data.
There is a growing need for the storage of such images on credit
cards and smart cards where the amount of storage available is
severely limited.  By combining a selective compression approach to
critical image features, such as EYES, NOSE and MOUTH, with a
particularly effective compression algorithm as described here, it is
possible to obtain substantial additional compression without
materially affecting  the quality of the image.  A portrait image
which would typically occupy  1500 bytes using waveform encoding can
be effectively compressed to about 250 bytes using the disclosed
technique.

      The compression algorithm used is the pyramid compression
algorithm described in detail in (*).  Briefly, in the conventional
pyramid image compression approach, a low resolution image is built
first by lowpass and then decimation.  The improved approach skips
the lowpass because the intervening pixels will be calculated anyway
so that there is no concern over aliasing effects.  The image is
decimated up to a distance of 16 pixels (starting with a portrait
image of 100X100 pixels).  At that distance, there is practically no
correlation left between adjacent pixels, therefore they may be
encoded as they are or at 6 bits per pixel.

      Next, every four neighboring pixels are used to predict the
pixel that lies in the geometrical centre, as in the following
scheme:
                          1  5  3  5  1
                          5  4  5  4  5
                          3  5  2  5  3
                          5  4  5  4  5
                          1  5  3  5  1

      Pixels named 1 (first level) are used to predict pixels named
2 (together they form the second level).  The second level is used to
predict pixels named 3 (together with 1 and 2 they form the third
level).  The third level is used to predict pixel...