UNIVERSITY OF JOENSUU DEPARTMENT OF COMPUTER SCIENCE Dictionarybased

UNIVERSITY OF JOENSUU DEPARTMENT OF COMPUTER SCIENCE Dictionary-based map image compression Image Compression Research group: http: //cs. joensuu. fi/pages/franti/comp/

The map service Real-time access to maps, independently from user location.

Properties of maps Maps of 5000 pixels (10 10 km 2). Uncompressed file size 12 Mb. Topographic and Road maps. Electronic library of Finnish Road maps with resolution 1: 250000 takes an entire CD (over 600 Mb) National Land Survey of Finland: www. nls. fi/index_e. html

Wireless communication GSM : 9600 bit per second. GPRS : 56 Kbit per second. Portable devices Small storage size 16/64 Mb. Weak processor performance: up to 200 Mhz.

Demands to map image compressor Quick and low machine-resource demandable decompressor

Semi-adaptive algorithms Negative properties Positive properties - Extra pass over the + Adaptive data + No updating of - Side information the model during needed compression

Block-segmentation of the image 1. Image blocks compressed separetly 2. Compressed blocks are stored in the same file 3. Header contains pointers to each block

LZW-based semi-adaptive compressin algorithm 1. Analyse of the image and built a global dictionary 2. Prune the dictionary 3. Compress image blocks separately

LZW algorithm Set word = NIL loop read a character k if word + k exists in the dictionary word = word + k else output the code for word add word+k to the dictionary w=k endloop

Step 1: The creation of the global dictionary 1. Divide the image into blocks 2. Process LZW encoding on each block 3. Keep the created global dictionary in the memory Principal scheme of the step 1

Step 2 : Pruning of the global dictionary 1. Encode the input image by the dictionary, created on the step 1. 2. Collect the statistics of the dictionary elements 3. Delete from the dictionary elements, which appeared during encoding less than two times

Step 3: The compression stage 1. Encode 2. Encode 3. Encode 4. Encode blocks the information about the image the pruned dictionary the image blocks pointers to the beginnings of the

Encoding of the image blocks 1. Modifiing of the cover elements statistics 2. Creating a Huffman tree 3. Encoding of the blocks

Decompression of the image 1. Read the header 2. Read the dictionary 3. Get access to the needed block 4. Decode the block

Experemental results. Compression raitios

Experemental results Dependency between the bit rate and the block size for PBM and PGM images.

Conclusions Good compression results (for dictionary-based compressors) Less influance for increasing of the block size