Multidisciplinary Pattern Recognition applications: A review

Marina Paolanti

doi:10.1016/j.cosrev.2020.100276

2020-08-01

PRE

Marina Paolanti *

Emanuele Frontoni

DOI：10.1016/j.cosrev.2020.100276

原文链接

原文求助

摘要

En 中文

Pattern recognition (PR) is the study of how machines can examine the environment, learn to distinguish patterns of interest from their background, and make reliable and feasible decisions regarding the categories of the patterns. However, even after almost 70 years of research, the design of an application based on pattern recognizer remains an ambiguous goal. Moreover, currently, there are huge volumes of data that must be dealt with, which include image, video, text and web documents; DNA; microarray gene data; etc. Among the various frameworks in which pattern recognition has been traditionally formulated, the statistical and machine learning approaches have been most comprehensively studied and employed in practice. Recently, deep learning techniques and methods have been receiving increasing attention. The main objective of this review is to summarize PR applications, departing from the major algorithms used for their design. The PR approaches are subdivided into three main methods: machine learning, statistical, and deep learning. In order to evidence the multidisciplinary aspects of PR applications, attention has been focused on latest PR methods applied to five fields of research: biomedical and biology, retail, surveillance, social media intelligence, and digital cultural heritage. In this paper, we discuss in detail the recent advances of PR approaches and propose the main applications within each field. We also present challenges and benchmarks in terms of advantages and disadvantages of the selected method in each field. A wide set of examples of applications in various domains are also provided, along with the specific method applied. (c) 2020 Elsevier Inc. All rights reserved.

Keyword:

Pattern recognition

Machine learning

Deep learning

Hidden Markov models