Research Document Imaging Storage System Architecture

Search

System Architecture and Features
of EDM Storage System

Storage System Requirements and Description

were outlined on previous page (EDM Storage System)

System Features

• Image is segmented into separate clusters of C ´ C pixels (eg. 128 ´ 128 pixels).
• Pointers (indices) to the clusters are stored to provide direct access to the clusters in the compressed file.
• Clusters are compressed with JBIG separately.
• Forward-adaptive context modelling is applied to reduce learning cost (small cluster problem):
  Two stage method:
  1. collect statistics over the image
  2. actual compression with re-initialisation
• Compression performance not worse than in JBIG:
  reduced learning cost compensates also the overhead of cluster indices and model tables
• Thumbnail consists of block codes:
  each pixel of preview image represents the 16 ´ 16 block in the original image classified either as black, white or mixed.
• Thumbnail takes < 2% of the compressed file and improves:
  1. compression ratio (about 2%, i.e. for FREE)
  2. compression/decompression time (~ 2.5 times)
• Once preview data, cluster indices, context models have been read in memory, any cluster can be decompressed separately.
• Only small modifications were needed for JBIG to achieve quick preview and spatial access to the image.

System Architecture Diagram

[click here to view]

Data organization of the compressed file

Main structure of the compression algorithm.

1. Analyze the image (analysis phase)
	1.1. Analyze block types 1.2. Construct the pixel level models
2. Write header and block level data
	2.1. Store text header 2.2. Compress block codes by QM-coder 2.3. Store dummy indexes 2.4. Store pixel level models
3. Process each cluster (compression phase)
	3.1. Compress the pixels by modified JBIG 3.2. Record the starting positions of the next clusters
4. End up compression
	4.1. Replace the dummy indexes by the real ones

Demonstration

Continue to EDM Storage System Demonstration page