Efficient algorithms for text and image compression based on knowledge engineering

Abstract

Huge volumes of data are generated in the World Wide Web in the form of text, image, audio and video etc. and there is an imminent need to compress them for efficient storage and transmission efficiency. In the past few decades, researchers have extensively worked in Data Compression, an active and challenging research area in the field of Information Theory. Data compression is a process that reduces the data size, removing the excessive information [62]. Naren Ramakrishnan et al. identified five different perspectives of Data Mining of which one is compression [140]. Data Mining is defined as the process of mining non-trivial, useful and hidden information from large databases. The process of Data Mining that focuses on generating a reduced (smaller) set of patterns (knowledge) from the original database can be viewed as a compression technique. The thesis presents a cluster of efficient and novel text and image compression algorithms using the perspective of Knowledge Engineering (Data Mining). We show some interesting highlights of the results obtained by our implementations with benchmark corpora by categorizing them into the following approaches. newline newlineFrequent Pattern Mining (FPM) based Huffman Encoding: Much of the text compression research has been focused on character/word based mechanism without looking at the larger perspective of pattern retrieval from dense datasets. We show how Frequent Pattern based Huffman encoding (FPH) employs Data Mining to efficiently compress text data. Mining the set of all frequent patterns from a database is an exponential task, not all of which is employed in the encoding process. An efficient FPM strategy to minimize the patterns required for encoding resulting in higher compression is presented.