Editing Compression Dictionaries toward Refined Compression-Based Feature-Space

This paper investigates how to construct a feature space for compression-based pattern recognition which judges the similarity between two objects x and y through the compression ratio to compress x with y (’s dictionary). Specifically, we focus on the known framework called PRDC, which represents a...

Full description

Saved in:
Bibliographic Details
Published inInformation (Basel) Vol. 13; no. 6; p. 301
Main Authors Koga, Hisashi, Ouchi, Shota, Nakajima, Yuji
Format Journal Article
LanguageEnglish
Published Basel MDPI AG 01.06.2022
Subjects
Algorithms
Clustering
compression dictionary
Compression ratio
compression-based pattern recognition
Data compression
Dictionaries
feature space
Image acquisition
Image classification
Object recognition
Pattern recognition
Support vector machines
Online AccessGet full text

Cover

More Information
Summary:This paper investigates how to construct a feature space for compression-based pattern recognition which judges the similarity between two objects x and y through the compression ratio to compress x with y (’s dictionary). Specifically, we focus on the known framework called PRDC, which represents an object x as a compression-ratio vector (CV) that lines up the compression ratios after x is compressed with multiple different dictionaries. By representing an object x as a CV, PRDC makes it possible to apply vector-based pattern recognition techniques to the compression-based pattern recognition. For PRDC, the dimensions, i.e., the dictionaries determine the quality of the CV space. This paper presents a practical technique to modify the chosen dictionaries in order to improve the performance of pattern recognition substantially: First, in order to make the dictionaries independent from each other, our method leaves any word shared by multiple dictionaries in only one dictionary and assures that any pair of dictionaries have no common words. Next, we transfer words among the dictionaries, so that all the dictionaries may keep roughly the same number of words and acquire the descriptive power evenly. The application to real image classification shows that our method increases classification accuracy by up to 8% compared with the case without our method, which demonstrates that our approach to keep the dictionaries independent is effective.
ISSN:2078-2489
2078-2489
DOI:10.3390/info13060301